You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

509 lines
20 KiB

Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Account domain blocks (#2381) * Add <ostatus:conversation /> tag to Atom input/output Only uses ref attribute (not href) because href would be the alternate link that's always included also. Creates new conversation for every non-reply status. Carries over conversation for every reply. Keeps remote URIs verbatim, generates local URIs on the fly like the rest of them. * Conversation muting - prevents notifications that reference a conversation (including replies, favourites, reblogs) from being created. API endpoints /api/v1/statuses/:id/mute and /api/v1/statuses/:id/unmute Currently no way to tell when a status/conversation is muted, so the web UI only has a "disable notifications" button, doesn't work as a toggle * Display "Dismiss notifications" on all statuses in notifications column, not just own * Add "muted" as a boolean attribute on statuses JSON For now always false on contained reblogs, since it's only relevant for statuses returned from the notifications endpoint, which are not nested Remove "Disable notifications" from detailed status view, since it's only relevant in the notifications column * Up max class length * Remove pending test for conversation mute * Add tests, clean up * Rename to "mute conversation" and "unmute conversation" * Raise validation error when trying to mute/unmute status without conversation * Adding account domain blocks that filter notifications and public timelines * Add tests for domain blocks in notifications, public timelines Filter reblogs of blocked domains from home * Add API for listing and creating account domain blocks * API for creating/deleting domain blocks, tests for Status#ancestors and Status#descendants, filter domain blocks from them * Filter domains in streaming API * Update account_domain_block_spec.rb
7 years ago
Account domain blocks (#2381) * Add <ostatus:conversation /> tag to Atom input/output Only uses ref attribute (not href) because href would be the alternate link that's always included also. Creates new conversation for every non-reply status. Carries over conversation for every reply. Keeps remote URIs verbatim, generates local URIs on the fly like the rest of them. * Conversation muting - prevents notifications that reference a conversation (including replies, favourites, reblogs) from being created. API endpoints /api/v1/statuses/:id/mute and /api/v1/statuses/:id/unmute Currently no way to tell when a status/conversation is muted, so the web UI only has a "disable notifications" button, doesn't work as a toggle * Display "Dismiss notifications" on all statuses in notifications column, not just own * Add "muted" as a boolean attribute on statuses JSON For now always false on contained reblogs, since it's only relevant for statuses returned from the notifications endpoint, which are not nested Remove "Disable notifications" from detailed status view, since it's only relevant in the notifications column * Up max class length * Remove pending test for conversation mute * Add tests, clean up * Rename to "mute conversation" and "unmute conversation" * Raise validation error when trying to mute/unmute status without conversation * Adding account domain blocks that filter notifications and public timelines * Add tests for domain blocks in notifications, public timelines Filter reblogs of blocked domains from home * Add API for listing and creating account domain blocks * API for creating/deleting domain blocks, tests for Status#ancestors and Status#descendants, filter domain blocks from them * Filter domains in streaming API * Update account_domain_block_spec.rb
7 years ago
Account domain blocks (#2381) * Add <ostatus:conversation /> tag to Atom input/output Only uses ref attribute (not href) because href would be the alternate link that's always included also. Creates new conversation for every non-reply status. Carries over conversation for every reply. Keeps remote URIs verbatim, generates local URIs on the fly like the rest of them. * Conversation muting - prevents notifications that reference a conversation (including replies, favourites, reblogs) from being created. API endpoints /api/v1/statuses/:id/mute and /api/v1/statuses/:id/unmute Currently no way to tell when a status/conversation is muted, so the web UI only has a "disable notifications" button, doesn't work as a toggle * Display "Dismiss notifications" on all statuses in notifications column, not just own * Add "muted" as a boolean attribute on statuses JSON For now always false on contained reblogs, since it's only relevant for statuses returned from the notifications endpoint, which are not nested Remove "Disable notifications" from detailed status view, since it's only relevant in the notifications column * Up max class length * Remove pending test for conversation mute * Add tests, clean up * Rename to "mute conversation" and "unmute conversation" * Raise validation error when trying to mute/unmute status without conversation * Adding account domain blocks that filter notifications and public timelines * Add tests for domain blocks in notifications, public timelines Filter reblogs of blocked domains from home * Add API for listing and creating account domain blocks * API for creating/deleting domain blocks, tests for Status#ancestors and Status#descendants, filter domain blocks from them * Filter domains in streaming API * Update account_domain_block_spec.rb
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
Non-Serial ("Snowflake") IDs (#4801) * Use non-serial IDs This change makes a number of nontrivial tweaks to the data model in Mastodon: * All IDs are now 8 byte integers (rather than mixed 4- and 8-byte) * IDs are now assigned as: * Top 6 bytes: millisecond-resolution time from epoch * Bottom 2 bytes: serial (within the millisecond) sequence number * See /lib/tasks/db.rake's `define_timestamp_id` for details, but note that the purpose of these changes is to make it difficult to determine the number of objects in a table from the ID of any object. * The Redis sorted set used for the feed will have values used to look up toots, rather than scores. This is almost always the same as the existing behavior, except in the case of boosted toots. This change was made because Redis stores scores as double-precision floats, which cannot store the new ID format exactly. Note that this doesn't cause problems with sorting/pagination, because ZREVRANGEBYSCORE sorts lexicographically when scores are tied. (This will still cause sorting issues when the ID gains a new significant digit, but that's extraordinarily uncommon.) Note a couple of tradeoffs have been made in this commit: * lib/tasks/db.rake is used to enforce many/most column constraints, because this commit seems likely to take a while to bring upstream. Enforcing a post-migrate hook is an easier way to maintain the code in the interim. * Boosted toots will appear in the timeline as many times as they have been boosted. This is a tradeoff due to the way the feed is saved in Redis at the moment, but will be handled by a future commit. This would effectively close Mastodon's #1059, as it is a snowflake-like system of generating IDs. However, given how involved the changes were simply within Mastodon, it may have unexpected interactions with some clients, if they store IDs as doubles (or as 4-byte integers). This was a problem that Twitter ran into with their "snowflake" transition, particularly in JavaScript clients that treated IDs as JS integers, rather than strings. It therefore would be useful to test these changes at least in the web interface and popular clients before pushing them to all users. * Fix JavaScript interface with long IDs Somewhat predictably, the JS interface handled IDs as numbers, which in JS are IEEE double-precision floats. This loses some precision when working with numbers as large as those generated by the new ID scheme, so we instead handle them here as strings. This is relatively simple, and doesn't appear to have caused any problems, but should definitely be tested more thoroughly than the built-in tests. Several days of use appear to support this working properly. BREAKING CHANGE: The major(!) change here is that IDs are now returned as strings by the REST endpoints, rather than as integers. In practice, relatively few changes were required to make the existing JS UI work with this change, but it will likely hit API clients pretty hard: it's an entirely different type to consume. (The one API client I tested, Tusky, handles this with no problems, however.) Twitter ran into this issue when introducing Snowflake IDs, and decided to instead introduce an `id_str` field in JSON responses. I have opted to *not* do that, and instead force all IDs to 64-bit integers represented by strings in one go. (I believe Twitter exacerbated their problem by rolling out the changes three times: once for statuses, once for DMs, and once for user IDs, as well as by leaving an integer ID value in JSON. As they said, "If you’re using the `id` field with JSON in a Javascript-related language, there is a very high likelihood that the integers will be silently munged by Javascript interpreters. In most cases, this will result in behavior such as being unable to load or delete a specific direct message, because the ID you're sending to the API is different than the actual identifier associated with the message." [1]) However, given that this is a significant change for API users, alternatives or a transition time may be appropriate. 1: https://blog.twitter.com/developer/en_us/a/2011/direct-messages-going-snowflake-on-sep-30-2011.html * Restructure feed pushes/unpushes This was necessary because the previous behavior used Redis zset scores to identify statuses, but those are IEEE double-precision floats, so we can't actually use them to identify all 64-bit IDs. However, it leaves the code in a much better state for refactoring reblog handling / coalescing. Feed-management code has been consolidated in FeedManager, including: * BatchedRemoveStatusService no longer directly manipulates feed zsets * RemoveStatusService no longer directly manipulates feed zsets * PrecomputeFeedService has moved its logic to FeedManager#populate_feed (PrecomputeFeedService largely made lots of calls to FeedManager, but didn't follow the normal adding-to-feed process.) This has the effect of unifying all of the feed push/unpush logic in FeedManager, making it much more tractable to update it in the future. Due to some additional checks that must be made during, for example, batch status removals, some Redis pipelining has been removed. It does not appear that this should cause significantly increased load, but if necessary, some optimizations are possible in batch cases. These were omitted in the pursuit of simplicity, but a batch_push and batch_unpush would be possible in the future. Tests were added to verify that pushes happen under expected conditions, and to verify reblog behavior (both on pushing and unpushing). In the case of unpushing, this includes testing behavior that currently leads to confusion such as Mastodon's #2817, but this codifies that the behavior is currently expected. * Rubocop fixes I could swear I made these changes already, but I must have lost them somewhere along the line. * Address review comments This addresses the first two comments from review of this feature: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336735 https://github.com/tootsuite/mastodon/pull/4801#discussion_r139336931 This adds an optional argument to FeedManager#key, the subtype of feed key to generate. It also tests to ensure that FeedManager's settings are such that reblogs won't be tracked forever. * Hardcode IdToBigints migration columns This addresses a comment during review: https://github.com/tootsuite/mastodon/pull/4801#discussion_r139337452 This means we'll need to make sure that all _id columns going forward are bigints, but that should happen automatically in most cases. * Additional fixes for stringified IDs in JSON These should be the last two. These were identified using eslint to try to identify any plain casts to JavaScript numbers. (Some such casts are legitimate, but these were not.) Adding the following to .eslintrc.yml will identify casts to numbers: ~~~ no-restricted-syntax: - warn - selector: UnaryExpression[operator='+'] > :not(Literal) message: Avoid the use of unary + - selector: CallExpression[callee.name='Number'] message: Casting with Number() may coerce string IDs to numbers ~~~ The remaining three casts appear legitimate: two casts to array indices, one in a server to turn an environment variable into a number. * Only implement timestamp IDs for Status IDs Per discussion in #4801, this is only being merged in for Status IDs at this point. We do this in a migration, as there is no longer use for a post-migration hook. We keep the initialization of the timestamp_id function as a Rake task, as it is also needed after db:schema:load (as db/schema.rb doesn't store Postgres functions). * Change internal streaming payloads to stringified IDs as well This is equivalent to 591a9af356faf2d5c7e66e3ec715502796c875cd from #5019, with an extra change for the addition to FeedManager#unpush. * Ensure we have a status_id_seq sequence Apparently this is not a given when specifying a custom ID function, so now we ensure it gets created. This uses the generic version of this function to more easily support adding additional tables with timestamp IDs in the future, although it would be possible to cut this down to a less generic version if necessary. It is only run during db:schema:load or the relevant migration, so the overhead is extraordinarily minimal. * Transition reblogs to new Redis format This provides a one-way migration to transition old Redis reblog entries into the new format, with a separate tracking entry for reblogs. It is not invertible because doing so could (if timestamp IDs are used) require a database query for each status in each users' feed, which is likely to be a significant toll on major instances. * Address review comments from @akihikodaki No functional changes. * Additional review changes * Heredoc cleanup * Run db:schema:load hooks for test in development This matches the behavior in Rails' ActiveRecord::Tasks::DatabaseTasks.each_current_configuration, which would otherwise break `rake db:setup` in development. It also moves some functionality out to a library, which will be a good place to put additional related functionality in the near future.
7 years ago
  1. require 'rails_helper'
  2. RSpec.describe FeedManager do
  3. before do |example|
  4. unless example.metadata[:skip_stub]
  5. stub_const 'FeedManager::MAX_ITEMS', 10
  6. stub_const 'FeedManager::REBLOG_FALLOFF', 4
  7. end
  8. end
  9. it 'tracks at least as many statuses as reblogs', skip_stub: true do
  10. expect(FeedManager::REBLOG_FALLOFF).to be <= FeedManager::MAX_ITEMS
  11. end
  12. describe '#key' do
  13. subject { FeedManager.instance.key(:home, 1) }
  14. it 'returns a string' do
  15. expect(subject).to be_a String
  16. end
  17. end
  18. describe '#filter?' do
  19. let(:alice) { Fabricate(:account, username: 'alice') }
  20. let(:bob) { Fabricate(:account, username: 'bob', domain: 'example.com') }
  21. let(:jeff) { Fabricate(:account, username: 'jeff') }
  22. context 'for home feed' do
  23. it 'returns false for followee\'s status' do
  24. status = Fabricate(:status, text: 'Hello world', account: alice)
  25. bob.follow!(alice)
  26. expect(FeedManager.instance.filter?(:home, status, bob.id)).to be false
  27. end
  28. it 'returns false for reblog by followee' do
  29. status = Fabricate(:status, text: 'Hello world', account: jeff)
  30. reblog = Fabricate(:status, reblog: status, account: alice)
  31. bob.follow!(alice)
  32. expect(FeedManager.instance.filter?(:home, reblog, bob.id)).to be false
  33. end
  34. it 'returns true for reblog by followee of blocked account' do
  35. status = Fabricate(:status, text: 'Hello world', account: jeff)
  36. reblog = Fabricate(:status, reblog: status, account: alice)
  37. bob.follow!(alice)
  38. bob.block!(jeff)
  39. expect(FeedManager.instance.filter?(:home, reblog, bob.id)).to be true
  40. end
  41. it 'returns true for reblog by followee of muted account' do
  42. status = Fabricate(:status, text: 'Hello world', account: jeff)
  43. reblog = Fabricate(:status, reblog: status, account: alice)
  44. bob.follow!(alice)
  45. bob.mute!(jeff)
  46. expect(FeedManager.instance.filter?(:home, reblog, bob.id)).to be true
  47. end
  48. it 'returns true for reblog by followee of someone who is blocking recipient' do
  49. status = Fabricate(:status, text: 'Hello world', account: jeff)
  50. reblog = Fabricate(:status, reblog: status, account: alice)
  51. bob.follow!(alice)
  52. jeff.block!(bob)
  53. expect(FeedManager.instance.filter?(:home, reblog, bob.id)).to be true
  54. end
  55. it 'returns true for reblog from account with reblogs disabled' do
  56. status = Fabricate(:status, text: 'Hello world', account: jeff)
  57. reblog = Fabricate(:status, reblog: status, account: alice)
  58. bob.follow!(alice, reblogs: false)
  59. expect(FeedManager.instance.filter?(:home, reblog, bob.id)).to be true
  60. end
  61. it 'returns false for reply by followee to another followee' do
  62. status = Fabricate(:status, text: 'Hello world', account: jeff)
  63. reply = Fabricate(:status, text: 'Nay', thread: status, account: alice)
  64. bob.follow!(alice)
  65. bob.follow!(jeff)
  66. expect(FeedManager.instance.filter?(:home, reply, bob.id)).to be false
  67. end
  68. it 'returns false for reply by followee to recipient' do
  69. status = Fabricate(:status, text: 'Hello world', account: bob)
  70. reply = Fabricate(:status, text: 'Nay', thread: status, account: alice)
  71. bob.follow!(alice)
  72. expect(FeedManager.instance.filter?(:home, reply, bob.id)).to be false
  73. end
  74. it 'returns false for reply by followee to self' do
  75. status = Fabricate(:status, text: 'Hello world', account: alice)
  76. reply = Fabricate(:status, text: 'Nay', thread: status, account: alice)
  77. bob.follow!(alice)
  78. expect(FeedManager.instance.filter?(:home, reply, bob.id)).to be false
  79. end
  80. it 'returns true for reply by followee to non-followed account' do
  81. status = Fabricate(:status, text: 'Hello world', account: jeff)
  82. reply = Fabricate(:status, text: 'Nay', thread: status, account: alice)
  83. bob.follow!(alice)
  84. expect(FeedManager.instance.filter?(:home, reply, bob.id)).to be true
  85. end
  86. it 'returns true for the second reply by followee to a non-federated status' do
  87. reply = Fabricate(:status, text: 'Reply 1', reply: true, account: alice)
  88. second_reply = Fabricate(:status, text: 'Reply 2', thread: reply, account: alice)
  89. bob.follow!(alice)
  90. expect(FeedManager.instance.filter?(:home, second_reply, bob.id)).to be true
  91. end
  92. it 'returns false for status by followee mentioning another account' do
  93. bob.follow!(alice)
  94. status = PostStatusService.new.call(alice, 'Hey @jeff')
  95. expect(FeedManager.instance.filter?(:home, status, bob.id)).to be false
  96. end
  97. it 'returns true for status by followee mentioning blocked account' do
  98. bob.block!(jeff)
  99. bob.follow!(alice)
  100. status = PostStatusService.new.call(alice, 'Hey @jeff')
  101. expect(FeedManager.instance.filter?(:home, status, bob.id)).to be true
  102. end
  103. it 'returns true for status by followee mentioning muted account' do
  104. bob.mute!(jeff)
  105. bob.follow!(alice)
  106. status = PostStatusService.new.call(alice, 'Hey @jeff')
  107. expect(FeedManager.instance.filter?(:home, status, bob.id)).to be true
  108. end
  109. it 'returns true for reblog of a personally blocked domain' do
  110. alice.block_domain!('example.com')
  111. alice.follow!(jeff)
  112. status = Fabricate(:status, text: 'Hello world', account: bob)
  113. reblog = Fabricate(:status, reblog: status, account: jeff)
  114. expect(FeedManager.instance.filter?(:home, reblog, alice.id)).to be true
  115. end
  116. context 'for irreversibly muted phrases' do
  117. it 'considers word boundaries when matching' do
  118. alice.custom_filters.create!(phrase: 'bob', context: %w(home), irreversible: true)
  119. alice.follow!(jeff)
  120. status = Fabricate(:status, text: 'bobcats', account: jeff)
  121. expect(FeedManager.instance.filter?(:home, status, alice.id)).to be_falsy
  122. end
  123. it 'returns true if phrase is contained' do
  124. alice.custom_filters.create!(phrase: 'farts', context: %w(home public), irreversible: true)
  125. alice.custom_filters.create!(phrase: 'pop tarts', context: %w(home), irreversible: true)
  126. alice.follow!(jeff)
  127. status = Fabricate(:status, text: 'i sure like POP TARts', account: jeff)
  128. expect(FeedManager.instance.filter?(:home, status, alice.id)).to be true
  129. end
  130. it 'matches substrings if whole_word is false' do
  131. alice.custom_filters.create!(phrase: 'take', context: %w(home), whole_word: false, irreversible: true)
  132. alice.follow!(jeff)
  133. status = Fabricate(:status, text: 'shiitake', account: jeff)
  134. expect(FeedManager.instance.filter?(:home, status, alice.id)).to be true
  135. end
  136. end
  137. end
  138. context 'for mentions feed' do
  139. it 'returns true for status that mentions blocked account' do
  140. bob.block!(jeff)
  141. status = PostStatusService.new.call(alice, 'Hey @jeff')
  142. expect(FeedManager.instance.filter?(:mentions, status, bob.id)).to be true
  143. end
  144. it 'returns true for status that replies to a blocked account' do
  145. status = Fabricate(:status, text: 'Hello world', account: jeff)
  146. reply = Fabricate(:status, text: 'Nay', thread: status, account: alice)
  147. bob.block!(jeff)
  148. expect(FeedManager.instance.filter?(:mentions, reply, bob.id)).to be true
  149. end
  150. it 'returns true for status by silenced account who recipient is not following' do
  151. status = Fabricate(:status, text: 'Hello world', account: alice)
  152. alice.update(silenced: true)
  153. expect(FeedManager.instance.filter?(:mentions, status, bob.id)).to be true
  154. end
  155. it 'returns false for status by followed silenced account' do
  156. status = Fabricate(:status, text: 'Hello world', account: alice)
  157. alice.update(silenced: true)
  158. bob.follow!(alice)
  159. expect(FeedManager.instance.filter?(:mentions, status, bob.id)).to be false
  160. end
  161. end
  162. end
  163. describe '#push_to_home' do
  164. it 'trims timelines if they will have more than FeedManager::MAX_ITEMS' do
  165. account = Fabricate(:account)
  166. status = Fabricate(:status)
  167. members = FeedManager::MAX_ITEMS.times.map { |count| [count, count] }
  168. Redis.current.zadd("feed:home:#{account.id}", members)
  169. FeedManager.instance.push_to_home(account, status)
  170. expect(Redis.current.zcard("feed:home:#{account.id}")).to eq FeedManager::MAX_ITEMS
  171. end
  172. context 'reblogs' do
  173. it 'saves reblogs of unseen statuses' do
  174. account = Fabricate(:account)
  175. reblogged = Fabricate(:status)
  176. reblog = Fabricate(:status, reblog: reblogged)
  177. expect(FeedManager.instance.push_to_home(account, reblog)).to be true
  178. end
  179. it 'does not save a new reblog of a recent status' do
  180. account = Fabricate(:account)
  181. reblogged = Fabricate(:status)
  182. reblog = Fabricate(:status, reblog: reblogged)
  183. FeedManager.instance.push_to_home(account, reblogged)
  184. expect(FeedManager.instance.push_to_home(account, reblog)).to be false
  185. end
  186. it 'saves a new reblog of an old status' do
  187. account = Fabricate(:account)
  188. reblogged = Fabricate(:status)
  189. reblog = Fabricate(:status, reblog: reblogged)
  190. FeedManager.instance.push_to_home(account, reblogged)
  191. # Fill the feed with intervening statuses
  192. FeedManager::REBLOG_FALLOFF.times do
  193. FeedManager.instance.push_to_home(account, Fabricate(:status))
  194. end
  195. expect(FeedManager.instance.push_to_home(account, reblog)).to be true
  196. end
  197. it 'does not save a new reblog of a recently-reblogged status' do
  198. account = Fabricate(:account)
  199. reblogged = Fabricate(:status)
  200. reblogs = 2.times.map { Fabricate(:status, reblog: reblogged) }
  201. # The first reblog will be accepted
  202. FeedManager.instance.push_to_home(account, reblogs.first)
  203. # The second reblog should be ignored
  204. expect(FeedManager.instance.push_to_home(account, reblogs.last)).to be false
  205. end
  206. it 'does not save a new reblog of a multiply-reblogged-then-unreblogged status' do
  207. account = Fabricate(:account)
  208. reblogged = Fabricate(:status)
  209. reblogs = 3.times.map { Fabricate(:status, reblog: reblogged) }
  210. # Accept the reblogs
  211. FeedManager.instance.push_to_home(account, reblogs[0])
  212. FeedManager.instance.push_to_home(account, reblogs[1])
  213. # Unreblog the first one
  214. FeedManager.instance.unpush_from_home(account, reblogs[0])
  215. # The last reblog should still be ignored
  216. expect(FeedManager.instance.push_to_home(account, reblogs.last)).to be false
  217. end
  218. it 'saves a new reblog of a long-ago-reblogged status' do
  219. account = Fabricate(:account)
  220. reblogged = Fabricate(:status)
  221. reblogs = 2.times.map { Fabricate(:status, reblog: reblogged) }
  222. # The first reblog will be accepted
  223. FeedManager.instance.push_to_home(account, reblogs.first)
  224. # Fill the feed with intervening statuses
  225. FeedManager::REBLOG_FALLOFF.times do
  226. FeedManager.instance.push_to_home(account, Fabricate(:status))
  227. end
  228. # The second reblog should also be accepted
  229. expect(FeedManager.instance.push_to_home(account, reblogs.last)).to be true
  230. end
  231. end
  232. it "does not push when the given status's reblog is already inserted" do
  233. account = Fabricate(:account)
  234. reblog = Fabricate(:status)
  235. status = Fabricate(:status, reblog: reblog)
  236. FeedManager.instance.push_to_home(account, status)
  237. expect(FeedManager.instance.push_to_home(account, reblog)).to eq false
  238. end
  239. end
  240. describe '#push_to_list' do
  241. let(:owner) { Fabricate(:account, username: 'owner') }
  242. let(:alice) { Fabricate(:account, username: 'alice') }
  243. let(:bob) { Fabricate(:account, username: 'bob') }
  244. let(:eve) { Fabricate(:account, username: 'eve') }
  245. let(:list) { Fabricate(:list, account: owner) }
  246. before do
  247. owner.follow!(alice)
  248. owner.follow!(bob)
  249. owner.follow!(eve)
  250. list.accounts << alice
  251. list.accounts << bob
  252. end
  253. it "does not push when the given status's reblog is already inserted" do
  254. reblog = Fabricate(:status)
  255. status = Fabricate(:status, reblog: reblog)
  256. FeedManager.instance.push_to_list(list, status)
  257. expect(FeedManager.instance.push_to_list(list, reblog)).to eq false
  258. end
  259. context 'when replies policy is set to no replies' do
  260. before do
  261. list.replies_policy = :no_replies
  262. end
  263. it 'pushes statuses that are not replies' do
  264. status = Fabricate(:status, text: 'Hello world', account: bob)
  265. expect(FeedManager.instance.push_to_list(list, status)).to eq true
  266. end
  267. it 'pushes statuses that are replies to list owner' do
  268. status = Fabricate(:status, text: 'Hello world', account: owner)
  269. reply = Fabricate(:status, text: 'Nay', thread: status, account: bob)
  270. expect(FeedManager.instance.push_to_list(list, reply)).to eq true
  271. end
  272. it 'does not push replies to another member of the list' do
  273. status = Fabricate(:status, text: 'Hello world', account: alice)
  274. reply = Fabricate(:status, text: 'Nay', thread: status, account: bob)
  275. expect(FeedManager.instance.push_to_list(list, reply)).to eq false
  276. end
  277. end
  278. context 'when replies policy is set to list-only replies' do
  279. before do
  280. list.replies_policy = :list_replies
  281. end
  282. it 'pushes statuses that are not replies' do
  283. status = Fabricate(:status, text: 'Hello world', account: bob)
  284. expect(FeedManager.instance.push_to_list(list, status)).to eq true
  285. end
  286. it 'pushes statuses that are replies to list owner' do
  287. status = Fabricate(:status, text: 'Hello world', account: owner)
  288. reply = Fabricate(:status, text: 'Nay', thread: status, account: bob)
  289. expect(FeedManager.instance.push_to_list(list, reply)).to eq true
  290. end
  291. it 'pushes replies to another member of the list' do
  292. status = Fabricate(:status, text: 'Hello world', account: alice)
  293. reply = Fabricate(:status, text: 'Nay', thread: status, account: bob)
  294. expect(FeedManager.instance.push_to_list(list, reply)).to eq true
  295. end
  296. it 'does not push replies to someone not a member of the list' do
  297. status = Fabricate(:status, text: 'Hello world', account: eve)
  298. reply = Fabricate(:status, text: 'Nay', thread: status, account: bob)
  299. expect(FeedManager.instance.push_to_list(list, reply)).to eq false
  300. end
  301. end
  302. context 'when replies policy is set to any reply' do
  303. before do
  304. list.replies_policy = :all_replies
  305. end
  306. it 'pushes statuses that are not replies' do
  307. status = Fabricate(:status, text: 'Hello world', account: bob)
  308. expect(FeedManager.instance.push_to_list(list, status)).to eq true
  309. end
  310. it 'pushes statuses that are replies to list owner' do
  311. status = Fabricate(:status, text: 'Hello world', account: owner)
  312. reply = Fabricate(:status, text: 'Nay', thread: status, account: bob)
  313. expect(FeedManager.instance.push_to_list(list, reply)).to eq true
  314. end
  315. it 'pushes replies to another member of the list' do
  316. status = Fabricate(:status, text: 'Hello world', account: alice)
  317. reply = Fabricate(:status, text: 'Nay', thread: status, account: bob)
  318. expect(FeedManager.instance.push_to_list(list, reply)).to eq true
  319. end
  320. it 'pushes replies to someone not a member of the list' do
  321. status = Fabricate(:status, text: 'Hello world', account: eve)
  322. reply = Fabricate(:status, text: 'Nay', thread: status, account: bob)
  323. expect(FeedManager.instance.push_to_list(list, reply)).to eq true
  324. end
  325. end
  326. end
  327. describe '#merge_into_timeline' do
  328. it "does not push source account's statuses whose reblogs are already inserted" do
  329. account = Fabricate(:account, id: 0)
  330. reblog = Fabricate(:status)
  331. status = Fabricate(:status, reblog: reblog)
  332. FeedManager.instance.push_to_home(account, status)
  333. FeedManager.instance.merge_into_timeline(account, reblog.account)
  334. expect(Redis.current.zscore("feed:home:0", reblog.id)).to eq nil
  335. end
  336. end
  337. describe '#trim' do
  338. let(:receiver) { Fabricate(:account) }
  339. it 'cleans up reblog tracking keys' do
  340. reblogged = Fabricate(:status)
  341. status = Fabricate(:status, reblog: reblogged)
  342. another_status = Fabricate(:status, reblog: reblogged)
  343. reblogs_key = FeedManager.instance.key('home', receiver.id, 'reblogs')
  344. reblog_set_key = FeedManager.instance.key('home', receiver.id, "reblogs:#{reblogged.id}")
  345. FeedManager.instance.push_to_home(receiver, status)
  346. FeedManager.instance.push_to_home(receiver, another_status)
  347. # We should have a tracking set and an entry in reblogs.
  348. expect(Redis.current.exists(reblog_set_key)).to be true
  349. expect(Redis.current.zrange(reblogs_key, 0, -1)).to eq [reblogged.id.to_s]
  350. # Push everything off the end of the feed.
  351. FeedManager::MAX_ITEMS.times do
  352. FeedManager.instance.push_to_home(receiver, Fabricate(:status))
  353. end
  354. # `trim` should be called automatically, but do it anyway, as
  355. # we're testing `trim`, not side effects of `push`.
  356. FeedManager.instance.trim('home', receiver.id)
  357. # We should not have any reblog tracking data.
  358. expect(Redis.current.exists(reblog_set_key)).to be false
  359. expect(Redis.current.zrange(reblogs_key, 0, -1)).to be_empty
  360. end
  361. end
  362. describe '#unpush' do
  363. let(:receiver) { Fabricate(:account) }
  364. it 'leaves a reblogged status if original was on feed' do
  365. reblogged = Fabricate(:status)
  366. status = Fabricate(:status, reblog: reblogged)
  367. FeedManager.instance.push_to_home(receiver, reblogged)
  368. FeedManager::REBLOG_FALLOFF.times { FeedManager.instance.push_to_home(receiver, Fabricate(:status)) }
  369. FeedManager.instance.push_to_home(receiver, status)
  370. # The reblogging status should show up under normal conditions.
  371. expect(Redis.current.zrange("feed:home:#{receiver.id}", 0, -1)).to include(status.id.to_s)
  372. FeedManager.instance.unpush_from_home(receiver, status)
  373. # Restore original status
  374. expect(Redis.current.zrange("feed:home:#{receiver.id}", 0, -1)).to_not include(status.id.to_s)
  375. expect(Redis.current.zrange("feed:home:#{receiver.id}", 0, -1)).to include(reblogged.id.to_s)
  376. end
  377. it 'removes a reblogged status if it was only reblogged once' do
  378. reblogged = Fabricate(:status)
  379. status = Fabricate(:status, reblog: reblogged)
  380. FeedManager.instance.push_to_home(receiver, status)
  381. # The reblogging status should show up under normal conditions.
  382. expect(Redis.current.zrange("feed:home:#{receiver.id}", 0, -1)).to eq [status.id.to_s]
  383. FeedManager.instance.unpush_from_home(receiver, status)
  384. expect(Redis.current.zrange("feed:home:#{receiver.id}", 0, -1)).to be_empty
  385. end
  386. it 'leaves a multiply-reblogged status if another reblog was in feed' do
  387. reblogged = Fabricate(:status)
  388. reblogs = 3.times.map { Fabricate(:status, reblog: reblogged) }
  389. reblogs.each do |reblog|
  390. FeedManager.instance.push_to_home(receiver, reblog)
  391. end
  392. # The reblogging status should show up under normal conditions.
  393. expect(Redis.current.zrange("feed:home:#{receiver.id}", 0, -1)).to eq [reblogs.first.id.to_s]
  394. reblogs[0...-1].each do |reblog|
  395. FeedManager.instance.unpush_from_home(receiver, reblog)
  396. end
  397. expect(Redis.current.zrange("feed:home:#{receiver.id}", 0, -1)).to eq [reblogs.last.id.to_s]
  398. end
  399. it 'sends push updates' do
  400. status = Fabricate(:status)
  401. FeedManager.instance.push_to_home(receiver, status)
  402. allow(Redis.current).to receive_messages(publish: nil)
  403. FeedManager.instance.unpush_from_home(receiver, status)
  404. deletion = Oj.dump(event: :delete, payload: status.id.to_s)
  405. expect(Redis.current).to have_received(:publish).with("timeline:#{receiver.id}", deletion)
  406. end
  407. end
  408. end