This pull request tries to optimize our status rethreading migration by using bulk updates + avoiding unnecessary writes, and doing the migration in one top-level loop and one stragglers loop, without the extra loop to copy thread_id over.
On my machine it runs at about 2400 rows per second on Postgres, now, and about 9000 rows per second on SQLite.
Tried *many* different ways of doing this, with and without temporary indexes, with different batch and transaction sizes, etc., and this seems to be just about the most performant way of getting stuff done.
With the changes, a few minutes have been shaved off migration time testing on my development machine. *Hopefully* this will translate to more time shaved off when running on a vps with slower read/write speed and less processor power.
SQLite before:
```
real 20m58,446s
user 16m26,635s
sys 5m53,648s
```
SQLite after:
```
real 14m25,435s
user 12m47,449s
sys 2m27,898s
```
Postgres before:
```
real 28m25,307s
user 3m40,005s
sys 4m45,018s
```
Postgres after:
```
real 22m31,999s
user 3m46,674s
sys 4m39,592s
```
Reviewed-on: https://codeberg.org/superseriousbusiness/gotosocial/pulls/4459
Co-authored-by: tobi <tobi.smethurst@protonmail.com>
Co-committed-by: tobi <tobi.smethurst@protonmail.com>
~~Still WIP!~~
This PR allows v0.20.0 of GtS to be forward-compatible with the interaction request / authorization flow that will fully replace the current flow in v0.21.0.
Basically, this means we need to recognize LikeRequest, ReplyRequest, and AnnounceRequest, and in response to those requests, deliver either a Reject or an Accept, with the latter pointing towards a LikeAuthorization, ReplyAuthorization, or AnnounceAuthorization, respectively. This can then be used by the remote instance to prove to third parties that the interaction has been accepted by the interactee. These Authorization types need to be dereferencable to third parties, so we need to serve them.
As well as recognizing the above "polite" interaction request types, we also need to still serve appropriate responses to "impolite" interaction request types, where an instance that's unaware of interaction policies tries to interact with a post by sending a reply, like, or boost directly, without wrapping it in a WhateverRequest type.
Doesn't fully close https://codeberg.org/superseriousbusiness/gotosocial/issues/4026 but gets damn near (just gotta update the federating with GtS documentation).
Migrations tested on both Postgres and SQLite.
Co-authored-by: kim <grufwub@gmail.com>
Reviewed-on: https://codeberg.org/superseriousbusiness/gotosocial/pulls/4394
Co-authored-by: tobi <tobi.smethurst@protonmail.com>
Co-committed-by: tobi <tobi.smethurst@protonmail.com>
# Description
This is quite a complex database migration that updates the `statuses.thread_id` column to be notnull, in order that statuses always be threaded, which will be useful in various pieces of upcoming work. This is unfortunately a migration that acts over the entire statuses table, and is quite complex in order to ensure that all existing statuses get correctly threaded together, and where possible fix any issues of statuses in the same thread having incorrect thread_ids.
TODO:
- ~~update testrig models to all be threaded~~
- ~~update code to ensure thread_id is always set~~
- ~~run on **a copy** of an sqlite production database~~
- ~~run on **a copy** of a postgres production database~~
## Checklist
- [x] I/we have read the [GoToSocial contribution guidelines](https://codeberg.org/superseriousbusiness/gotosocial/src/branch/main/CONTRIBUTING.md).
- [x] I/we have discussed the proposed changes already, either in an issue on the repository, or in the Matrix chat.
- [x] I/we have not leveraged AI to create the proposed changes.
- [x] I/we have performed a self-review of added code.
- [x] I/we have written code that is legible and maintainable by others.
- [x] I/we have commented the added code, particularly in hard-to-understand areas.
- [ ] I/we have made any necessary changes to documentation.
- [x] I/we have added tests that cover new code.
- [x] I/we have run tests and they pass locally with the changes.
- [x] I/we have run `go fmt ./...` and `golangci-lint run`.
Reviewed-on: https://codeberg.org/superseriousbusiness/gotosocial/pulls/4160
Co-authored-by: kim <grufwub@gmail.com>
Co-committed-by: kim <grufwub@gmail.com>
* [feature] Parse content warning as HTML, serialize via API to plaintext
* tidy up some cruft
* whoops
* oops
* i'm da joker baybee
* clemency muy lorde
* rename some of the text functions for clarity
* jiggle the opts
* fiddle de deee
* hopefully the last test fix i ever have to do in my beautiful life
* Add ContentType to internal models
* Add ContentType to API models StatusSource and StatusEdit
* Add helpers to convert between API/internal StatusContentType
* Write status content type on create/edit
* Add migration
* Update API docs
go run github.com/go-swagger/go-swagger/cmd/swagger generate spec --scan-models --exclude-deps --output docs/api/swagger.yaml
* ensure ContentType is updated anywhere Text is
* Update docs, take care of TODOs
* Set ContentType in more places where Text is set
* We don't actually use ContentType on the API status model
* Update StatusSource test
* Remove unused helper function I copied
* Revert change to StatusContentType swagger annotation
I'm going to include this in a follow-on PR instead.
* Add test for updating content type in edits
* Return a value from processContentType instead of modifying the existing status
Fixes an issue that was caught by the test I just added - the recorded edit would be marked with the *new* content type instead of the old one, which is obviously bad
* Add test for handling of statuses with no stored content type
* repurpose an existing test status instead of adding a new one to avoid breaking other tests
* Add test to ensure newly created statuses always have content type saved
* Do include content type on status API model actually
This is mostly important when deleting and redrafting.
The comment on `apimodel.Status.Text` implies that it's not sent except in response to status deletion, but actually this doesn't seem to be the case; it also appears to be present in responses to creations and normal fetches and stuff. So I'm treating `ContentType` the same here.
* Update new tests to check content type on API statuses
* Check content type of API statuses in all tests where text is checked
* update other api tests with status content type field
* Add test ensuring text and content type are returned when deleting a status
* Convert processContentType to free function and remove unused parameter
* check for the correct value in the deletion test
* Be explicit about this test status having an empty content type
* Use omitempty consistently on API models
* clean up the final diff a bit
* one more swagger regen for the road
* Handle nil statuses in processContentType
* Don't pass processContentType the entire edit form, it doesn't need it
* Move processContentType to common.go and use for creation as well
* Remove unused parameters to ContentTypeToAPIContentType
* Implement backfilling statuses thru scheduled_at
* Forbid mentioning others in backfills
* Update error messages & codes
* Add new tests for backfilled statuses
* Test that backfilling doesn't timeline or notify
* Fix check for absence of notification
* Test that backfills do not cause federation
* Fix type of apimodel.StatusCreateRequest.ScheduledAt in tests
* Add config file switch and min date check
* update statuses table to replace updated_at column with edited_at
* code comment
* better code comments, fix setting of status + edit + mention + poll database times
* fix log to logf call
* fix status.EditIDs not being carried over in dereferencer.encrichStatus()
* move status.EditID setting into handleStatusEdit()
* add support for extracting Updated field from Statusable implementers
* add support for status edits in the database, and update status dereferencer to handle them
* remove unused AdditionalInfo{}.CreatedAt
* remove unused AdditionalEmojiInfo{}.CreatedAt
* update new mention creation to use status.UpdatedAt
* remove mention.UpdatedAt, fixes related to NewULIDFromTime() change
* add migration to remove Mention{}.UpdatedAt field
* add migration to add the StatusEdit{} table
* start adding tests, add delete function for status edits
* add more of status edit migrations, fill in more of the necessary edit delete functionality
* remove unused function
* allow generating gotosocial compatible ulid via CLI with `go run ./cmd/gen-ulid`
* add StatusEdit{} test models
* fix new statusedits sql
* use model instead of table name
* actually remove the Mention.UpdatedAt field...
* fix tests now new models are added, add more status edit DB tests
* fix panic wording
* add test for deleting status edits
* don't automatically set `updated_at` field on updated statuses
* flesh out more of the dereferencer status edit tests, ensure updated at field set on outgoing AS statuses
* remove media_attachments.updated_at column
* fix up more tests, further complete the dereferencer status edit tests
* update more status serialization tests not expecting 'updated' AS property
* gah!! json serialization tests!!
* undo some gtscontext wrapping changes
* more serialization test fixing 🥲
* more test fixing, ensure the edit.status_id field is actually set 🤦
* fix status edit test
* grrr linter
* add edited_at field to apimodel status
* remove the choice of paging on the timeline public filtered test (otherwise it needs updating every time you add statuses ...)
* ensure that status.updated_at always fits chronologically
* fix more serialization tests ...
* add more code comments
* fix envparsing
* update swagger file
* properly handle media description changes during status edits
* slight formatting tweak
* code comment
* convert statuses.visibility and notifications.notification_type columns from type string -> int for performance / space savings
* fix test trying to compare string to int
* fix instance count query using string literal instead of gtsmodel const type
* ensure a default value is always set
* also migrate the account settings and sin bin status tables
* initialize maps outside loops and place into singular enum mapping creation func
* use int16 for enum types
* update sinbinstatus creation to be from a snapshot at initial creation
* add snapshot of poll type at creation time
* search for mentions also by username,domain in status deref, handle deleted statuses in enrichStatusSafely()
* return d.enrichStatusSafely() directly
* [feature/frontend] Add options to include Unlisted posts or hide all posts
* finish up
* swagger
* move invalidate call into bundb package, avoid invalidating if not necessary
* rename show_web_statuses => web_visibility
* don't use ptr for webvisibility
* last bits
* [feature] Federate interaction policies + Accepts; enforce policies
* use Acceptable type
* fix index
* remove appendIRIStrs
* add GetAccept federatingdb function
* lock on object IRI
* [feature] Federate status language in + out
* go fmt
* tests, little fix
* improve comments
* unnest a bit
* avoid unnecessary nil check
* use more descriptive variable for contentMap
* prefer instance languages when selecting from contentMap
* update docs to reflect lang selection
* rename rdfLangString -> rdfLangs
* update comments to mention Pollable
* iter through slice instead of map
* add db models + functions for keeping track of threads
* give em the old linty testy
* create, remove, check mutes
* swagger
* testerino
* test mute/unmute via api
* add info log about new index creation
* thread + allow muting of any remote statuses that mention a local account
* IsStatusThreadMutedBy -> IsThreadMutedByAccount
* use common processing functions in status processor
* set = NULL
* favee!
* get rekt darlings, darlings get rekt
* testrig please, have mercy muy liege
* [chore] Remove go-playground/validator
It turns out we're not actually using the validator code. This is a
remnant from when we intended to use it, but the presence of it and its
struct tags creates the illusion we're validating a lot of things we're
not. It resulted in some confusion when we were trying to figure out
language valdiation.
Remove all this code, so that only the validation functions from the
validate package we actually use remain. I'm not touching the struct
tags in the migrations in order to avoid things potentially thinking
migrations need to be re-run.
* [chore] Bring back a struct tag on api
The validate on internal/api is Gin doing form validation, not the
validator from go-playground/validator.
* revamp http client to not limit requests, instead use sender worker
Signed-off-by: kim <grufwub@gmail.com>
* remove separate sender worker pool, spawn 2*GOMAXPROCS batch senders each time, no need for transport cache sweeping
Signed-off-by: kim <grufwub@gmail.com>
* improve batch senders to keep popping recipients until remote URL found
Signed-off-by: kim <grufwub@gmail.com>
* fix recipient looping issue
Signed-off-by: kim <grufwub@gmail.com>
* move request id ctx key to gtscontext, finish filling out more code comments, add basic support for not logging client IP
Signed-off-by: kim <grufwub@gmail.com>
* first draft of status refetching logic
Signed-off-by: kim <grufwub@gmail.com>
* fix testrig to use new federation alloc func signature
Signed-off-by: kim <grufwub@gmail.com>
* fix log format directive
Signed-off-by: kim <grufwub@gmail.com>
* add status fetched_at migration
Signed-off-by: kim <grufwub@gmail.com>
* remove unused / unchecked for error types
Signed-off-by: kim <grufwub@gmail.com>
* add back the used type...
Signed-off-by: kim <grufwub@gmail.com>
* add separate internal getStatus() function for derefThread() that doesn't recurse
Signed-off-by: kim <grufwub@gmail.com>
* improved mention and media attachment error handling
Signed-off-by: kim <grufwub@gmail.com>
* fix log and error format directives
Signed-off-by: kim <grufwub@gmail.com>
* update account deref to match status deref changes
Signed-off-by: kim <grufwub@gmail.com>
* very small code formatting change to make things clearer
Signed-off-by: kim <grufwub@gmail.com>
* add more code comments
Signed-off-by: kim <grufwub@gmail.com>
* improved code commenting
Signed-off-by: kim <grufwub@gmail.com>
* only check for required further derefs if needed
Signed-off-by: kim <grufwub@gmail.com>
* improved cache invalidation
Signed-off-by: kim <grufwub@gmail.com>
* tweak cache restarting to use a (very small) backoff
Signed-off-by: kim <grufwub@gmail.com>
* small readability changes and fixes
Signed-off-by: kim <grufwub@gmail.com>
* fix account sync issues
Signed-off-by: kim <grufwub@gmail.com>
* fix merge conflicts + update account enrichment to accept already-passed accountable
Signed-off-by: kim <grufwub@gmail.com>
* remove secondary function declaration
Signed-off-by: kim <grufwub@gmail.com>
* normalise dereferencer get status / account behaviour, fix remaining tests
Signed-off-by: kim <grufwub@gmail.com>
* fix remaining rebase conflicts, finish commenting code
Signed-off-by: kim <grufwub@gmail.com>
* appease the linter
Signed-off-by: kim <grufwub@gmail.com>
* add source file header
Signed-off-by: kim <grufwub@gmail.com>
* update to use TIMESTAMPTZ column type instead of just TIMESTAMP
Signed-off-by: kim <grufwub@gmail.com>
* don't pass in 'updated_at' to UpdateEmoji()
Signed-off-by: kim <grufwub@gmail.com>
* use new ap.Resolve{Account,Status}able() functions
Signed-off-by: kim <grufwub@gmail.com>
* remove the somewhat confusing rescoping of the same variable names
Signed-off-by: kim <grufwub@gmail.com>
* update migration file name, improved database delete error returns
Signed-off-by: kim <grufwub@gmail.com>
* formatting
Signed-off-by: kim <grufwub@gmail.com>
* improved multi-delete database functions to minimise DB calls
Signed-off-by: kim <grufwub@gmail.com>
* remove unused type
Signed-off-by: kim <grufwub@gmail.com>
* fix delete statements
Signed-off-by: kim <grufwub@gmail.com>
---------
Signed-off-by: kim <grufwub@gmail.com>
* refactor visibility checking, add caching for visibility
* invalidate visibility cache items on account / status deletes
* fix requester ID passed to visibility cache nil ptr
* de-interface caches, fix home / public timeline caching + visibility
* finish adding code comments for visibility filter
* fix angry goconst linter warnings
* actually finish adding filter visibility code comments for timeline functions
* move home timeline status author check to after visibility
* remove now-unused code
* add more code comments
* add TODO code comment, update printed cache start names
* update printed cache names on stop
* start adding separate follow(request) delete db functions, add specific visibility cache tests
* add relationship type caching
* fix getting local account follows / followed-bys, other small codebase improvements
* simplify invalidation using cache hooks, add more GetAccountBy___() functions
* fix boosting to return 404 if not boostable but no error (to not leak status ID)
* remove dead code
* improved placement of cache invalidation
* update license headers
* add example follow, follow-request config entries
* add example visibility cache configuration to config file
* use specific PutFollowRequest() instead of just Put()
* add tests for all GetAccountBy()
* add GetBlockBy() tests
* update block to check primitive fields
* update and finish adding Get{Account,Block,Follow,FollowRequest}By() tests
* fix copy-pasted code
* update envparsing test
* whitespace
* fix bun struct tag
* add license header to gtscontext
* fix old license header
* improved error creation to not use fmt.Errorf() when not needed
* fix various rebase conflicts, fix account test
* remove commented-out code, fix-up mention caching
* fix mention select bun statement
* ensure mention target account populated, pass in context to customrenderer logging
* remove more uncommented code, fix typeutil test
* add statusfave database model caching
* add status fave cache configuration
* add status fave cache example config
* woops, catch missed error. nice catch linter!
* add back testrig panic on nil db
* update example configuration to match defaults, slight tweak to cache configuration defaults
* update envparsing test with new defaults
* fetch followingget to use the follow target account
* use accounnt.IsLocal() instead of empty domain check
* use constants for the cache visibility type check
* use bun.In() for notification type restriction in db query
* include replies when fetching PublicTimeline() (to account for single-author threads in Visibility{}.StatusPublicTimelineable())
* use bun query building for nested select statements to ensure working with postgres
* update public timeline future status checks to match visibility filter
* same as previous, for home timeline
* update public timeline tests to dynamically check for appropriate statuses
* migrate accounts to allow unique constraint on public_key
* provide minimal account with publicKey
---------
Signed-off-by: kim <grufwub@gmail.com>
Co-authored-by: tsmethurst <tobi.smethurst@protonmail.com>
* [chore] Remove years from all license headers
Years or year ranges aren't required in license headers. Many projects
have removed them in recent years and it avoids a bit of yearly toil.
In many cases our copyright claim was also a bit dodgy since we added
the 2021-2023 header to files created after 2021 but you can't claim
copyright into the past that way.
* [chore] Add license header check
This ensures a license header is always added to any new file. This
avoids maintainers/reviewers needing to remember to check for and ask
for it in case a contribution doesn't include it.
* [chore] Add missing license headers
* [chore] Further updates to license header
* Use the more common // indentend comment format
* Remove the hack we had for the linter now that we use the // format
* Add SPDX license identifier
* implement status pin client api + web handler
* make test names + comments more descriptive
* don't use separate table for status pins
* remove unused add + remove checking
* tidy up + add some more tests
* upstep bun and sqlite versions
* allow specific columns to be updated in the db
* only update necessary columns for user
* bit tidier
* only update necessary fields of media_attachment
* only update relevant instance fields
* update tests
* update only specific account columns
* use bool pointers on gtsmodels
includes attachment, status, account, user
* update columns more selectively
* test all default fields on new account insert
* updating remaining bools on gtsmodels
* initialize pointer fields when extracting AP emoji
* copy bools properly
* add copyBoolPtr convenience function + test it
* initialize false bool ptrs a bit more neatly
* add preparable and timelineable interfaces
* initialize timeline manager within the processor
* generic renaming
* move status-specific timeline logic into the processor
* refactor timeline to make it useful for more than statuses
* prune gtsmodel.Account bun tags, add note to gtsmodel dir
Signed-off-by: kim (grufwub) <grufwub@gmail.com>
* further database tag fixes
Signed-off-by: kim (grufwub) <grufwub@gmail.com>
* more db tag fixups
Signed-off-by: kim (grufwub) <grufwub@gmail.com>
* fix removing nullzero for account timestamps...
Signed-off-by: kim (grufwub) <grufwub@gmail.com>
* add nullzero back to accountid tag
Signed-off-by: kim (grufwub) <grufwub@gmail.com>
* rename gtsmodel readme
Signed-off-by: kim (grufwub) <grufwub@gmail.com>
* start moving some database stuff around
* continue moving db stuff around
* more fiddling
* more updates
* and some more
* and yet more
* i broke SOMETHING but what, it's a mystery
* tidy up
* vendor ttlcache
* use ttlcache
* fix up some tests
* rename some stuff
* little reminder
* some more updates
* decided where to put reply dereferencing
* fiddling with dereferencing threads
* further adventures
* tidy up some stuff
* move dereferencing functionality
* a bunch of refactoring
* go fmt
* more refactoring
* bleep bloop
* docs and linting
* start implementing replies collection on gts side
* fiddling around
* allow dereferencing our replies
* lint, fmt
* start messing about with timeline manager
* i have no idea what i'm doing
* i continue to not know what i'm doing
* it's coming along
* bit more progress
* update timeline with new posts as they come in
* lint and fmt
* Select accounts where empty string
* restructure a bunch, get unfaves working
* moving stuff around
* federate status deletes properly
* mention regex better but not 100% there
* fix regex
* some more hacking away at the timeline code phew
* fix up some little things
* i can't even
* more timeline stuff
* move to ulid
* fiddley
* some lil fixes for kibou compatibility
* timelines working pretty alright!
* tidy + lint
