Gun's connection pool also returns an error if duplicate workers are launched simultaneously. Snooze on this error as well, and lower the snooze to 3 seconds with the optimism that the connection will still be open by then and the delivery can be completed quickly.
The original setting of 30 seconds is pretty high and means there's an unnatural lag between deliveries of activities destined to the same server that were created at nearly the same time. This configuration should be more efficient.
Our test environment cheats by constructing a conn with a custom oauth_access/2 function. This assigns a :token to the conn but due to the way it is constructed it has the :user preloaded. When the OAuth Plug fetches a token it does not preload the user, so the check for user.disclose_client was always nil and assumed to be false.
Preloading the :user ensures the test environment matches reality.
It is angry we are making a fake %Plug.Conn{} to pass through Signature.validate_signature/1. We can work around it by making the code support a map, but then we lose the benefit of being able to use put_req_header/3
This logic only exists in the Plug, so attempting to validate the signature by calling the library function HTTPSignature.validate_conn/2 directly will never work because we do not attempt to construct the (request-target) and @request-target headers with both the commonly misinterpreted and correct implementation of this field. Therefore all attempts to validate a signature from an Oban Job will fail.
When signatures fail on incoming activities we put the job into Oban to be processed later instead of doing the user fetching and validation inline which is expensive and increases latency on the incoming POST request. Unfortunately we did not retain the :method, :request_path, and :query_string parameters from the conn so the signature validation and Oban Job would always fail.
This was most obvious when Mastodon sends Deletes for users your server has never seen before.
This is for a normal HTTP error response or timeout while receiving the data. A hard error from a process crash, DNS lookup failure, etc should produce a different response than {:ok, %Tesla.Env{}} and the request/job will be retryable.
This should fix WithClauseError resulting in Oban jobs for processing
incoming deletes being retried without getting cancelled when those
deletes are MRF rejected.
Fixes module name being not fully qualified
warning: AdminAPI.FallbackController.call/2 is undefined (module AdminAPI.FallbackController is not available or is yet to be defined)
│
5 │ defmodule Pleroma.Web.AdminAPI.RuleController do
│ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
│
└─ lib/pleroma/web/admin_api/controllers/rule_controller.ex:5: Pleroma.Web.AdminAPI.RuleController.action/2
warning: AdminAPI.FallbackController.init/1 is undefined (module AdminAPI.FallbackController is not available or is yet to be defined)
│
5 │ defmodule Pleroma.Web.AdminAPI.RuleController do
│ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
│
└─ lib/pleroma/web/admin_api/controllers/rule_controller.ex:5: Pleroma.Web.AdminAPI.RuleController.action/2
Key the cache on the image being used and the commit sha. This should allow the cache to be reused by the same runner across multiple jobs/stages where appropriate.
Also update other places where we use the term "send" instead of "stream". This should make it clearer that we are streaming these over websockets / web push and not sending an activity.
:idna.encode/1 expects a charlist even though it will accept a binary string. That functionality is undocumented / not part of its typespec, so we should turn it into a charlist first. Also switch to using match?/2
lib/pleroma/user.ex:2056:call
The function call will not succeed.
:idna.encode(_host :: binary())
will never return since the success typing is:
(string()) :: string()
and the contract is
(string()) :: string()
Wrong @spec name for remove_from_block/2
lib/pleroma/user.ex:2721:overlapping_contract
Overloaded contract for Pleroma.User.add_to_block/2 has
overlapping domains; such contracts are currently unsupported and
are simply ignored.
WebPushEncryption.send_web_push/4 was written to raise on erroroneus input, so we must guard against that.
lib/pleroma/web/push/impl.ex:65:no_return Function push_message/4 has no local return.
lib/pleroma/web/media_proxy/media_proxy_controller.ex:154:pattern_match
The pattern can never match the type.
Pattern:
{:ok, _thumbnail_binary}
Type:
{:error, boolean() | {:ffmpeg, :command_not_found}}
The callback already defines the @spec and these do not match it.
lib/pleroma/upload/filter/exiftool/strip_location.ex:12:callback_spec_type_mismatch
The @spec return type does not match the expected return type
for filter/1 callback in Pleroma.Upload.Filter behaviour.
Actual:
@spec filter(...) :: {:ok, _}
Expected:
@spec filter(...) :: {:error, _} | {:ok, :filtered | :noop} | {:ok, :filtered, struct()}
lib/pleroma/notification.ex:492:invalid_contract
The @spec for the function does not match the success typing of the function.
Function:
Pleroma.Notification.get_notified_from_activity/2
Success typing:
@spec get_notified_from_activity(_, _) :: [any()]
Sharing this pool with regular Media is problematic as Rich Media will connect to many different
domains and thrash the pool, but regular Media will have predictable connections to the webservers
hosting media for the fediverse servers you peer with.
From realpath(1) in POSIX 202x Draft 4.1:
> If file does not name a symbolic link, readlink shall write a diagnostic
> message to standard error and exit with non-zero status.
Which also doesn't includes `-f`, in preference of `realpath`.
@ -4,6 +4,159 @@ All notable changes to this project will be documented in this file.
The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
## 2.7.0
### Security
- HTTP Security: By default, don't allow unsafe-eval. The setting needs to be changed to allow Flash emulation.
- Fix webfinger spoofing.
- Use proper workers for fetching pins instead of an ad-hoc task, fixing a potential fetch loop
### Changed
- Update to Phoenix 1.7
- Elixir Logger configuration is now longer permitted through AdminFE and ConfigDB
- Refactor the user backups code and improve test coverage
- Invalid activities delivered to the inbox will be rejected with a 400 Bad Request
- Support Bandit as an alternative to Cowboy for the HTTP server.
- Update Bandit to 1.5.2
- Replace eblurhash with rinpatch_blurhash. This also removes a dependency on ImageMagick.
- Elixir 1.13 is the minimum required version.
- Document maximum supported version of Erlang & Elixir
- Update and extend NetBSD installation docs
- Make `/api/v1/pleroma/federation_status` publicly available
- Increase outgoing federation parallelism
- Change Hackney connection pool timeouts to align with the values Gun uses
- Transmogrifier: handle non-validate errors on incoming Delete activities
- Remote object fetch failures will prevent the object fetch job from retrying if the object request returns 401, 403, 404, 410, or exceeds the maximum thread depth.
- - Change AccountView `last_status_at` from a datetime to a date (as done in Mastodon 3.1.0)
- Improve error logging when LDAP authentication fails.
- Publisher jobs will not retry if the error received is a 400
- PollWorker jobs will not retry if the activity no longer exists.
- Improved detecting unrecoverable errors for incoming federation jobs
- Changed some jobs to return :cancel on unrecoverable errors that should not be retried
- Discard Remote Fetcher jobs which errored due to an MRF rejection.
- Oban queues have refactored to simplify the queue design
- Ensure all Oban jobs have timeouts defined
- Optimistic Inbox reduces the processing overhead of incoming activities without instantly verifiable signatures.
- HTTP connection pool adjustments
- Disable jit by default for PostgreSQL
- Update the documentation for configuring Prometheus metrics.
- Change the prometheus library to PromEx.
- Publisher jobs now store the the activity id instead of inserting duplicate JSON data in the Oban queue for each delivery.
- Activity publishing failures will prevent the job from retrying if the publishing request returns a 403 or 410
- Publisher errors will now emit logs indicating the inbox that was not available for delivery.
- Reduce the reachability timestamp update to a single upsert query
- A 422 error is returned when attempting to reply to a deleted status
- Rich Media backfilling is now an Oban job
- Refactored Rich Media to cache the content in the database. Fetching operations that could block status rendering have been eliminated.
- Set default values on validators for transient objects (attachment, poll options)
- User profile refreshes are now asynchronous
- Change mediaproxy previews to use vips to generate thumbnails instead of ImageMagick
- Render nice web push notifications for polls
- Refactor the Mastodon /api/v1/streaming websocket handler to use Phoenix.Socket.Transport
### Added
- Uploader: Add support for uploading attachments using IPFS
- Add NSFW-detecting MRF
- Add DNSRBL MRF
- Add options to the mix prune_objects task
- Add Anti-mention Spam MRF backported from Rebased
- Support /authorize-interaction route used by Mastodon
- Add an option to reject certain domains when authorized fetch is enabled.
- Include following/followers in backups
- Allow to group bookmarks in folders
- Include image description in status media cards
- Implement `/api/v1/accounts/familiar_followers`
- Add support for configuring favicon, embed favicon and PWA manifest in server-generated meta
- Implement FEP-2c59, add "webfinger" to user actor
- Framegrabs with ffmpeg will execute with a 5 second timeout and cache the URLs of failures with a TTL of 15 minutes to prevent excessive retries.
- Added a Mix task "pleroma.config fix_mrf_policies" which will remove erroneous MRF policies from ConfigDB.
- Add ForceMention MRF
- [docs] add frontends management documentation
- Implement group actors
- Add contact account to InstanceView
- Add instance rules
- Implement /api/v2/instance route
- Verify profile link ownership with rel="me"
- Logger metadata is now attached to some logs to help with troubleshooting and analysis
- Add new parameters to /api/v2/instance: configuration[accounts][max_pinned_statuses] and configuration[statuses][characters_reserved_per_url]
- Add meilisearch, make search engines pluggable
- Add missing indexes on foreign key relationships
- Startup detection for configured MRF modules that are missing or incorrectly defined
- Permit passing --chunk and --step values to the Pleroma.Search.Indexer Mix task
- Deleting, Unfavoriting, Unrepeating, or Unreacting will cancel undelivered publishing jobs for the original activity.
- Oban jobs can now be viewed in the Live Dashboard
- Add media proxy to opengraph rich media cards
- Support for Erlang OTP 26
- Prioritize mentioned recipients (i.e., those that are not just followers) when federating.
- PromEx documentation
- Expose nonAnonymous field from Smithereen polls
- Add Qdrant/OpenAI embedding search
- Adds the capability to add a URL to a scrobble (optional field)
- scrubbers/default: Add more formatting elements from HTML4 / GoToSocial (acronym, bdo, big, cite, dfn, ins, kbd, q, samp, s, tt, var, wbr)
- Monitoring of search backend health to control the processing of jobs in the search indexing Oban queue
- Display reposted replies with exclude_replies: true
- Add "status" notification type
- Support honk-style attachment summaries as alt-text.
### Fixed
- Fix Emoji object IDs not always being valid
- Remove checking ImageMagick's commands for Pleroma.Upload.Filter.AnalyzeMetadata
- Ensure that StripLocation actually removes everything resembling GPS data from PNGs
- Fix authentication check on account rendering when bio is defined
- ap userview: add outbox field.
- Fix #strip_report_status_data
- Fix federation with Convergence AP Bridge
- ChatMessage: Tolerate attachment field set to an empty array
- Config: Check the permissions of the linked file instead of the symlink
- MediaProxy was setting the content-length header which is not permitted by RFC9112§6.2 when we are chunking the reply as it conflicts with the existence of the transfer-encoding header.
- Restore Cowboy's ability to stream MediaProxy responses without Chunked encoding.
- Fix the processing of email digest jobs.
- Client application data was always missing from the status
- Elixir 1.15 compatibility
- When downloading remote emojis packs, account for pagination
- Make remote emoji packs API use specifically the V1 URL. Akkoma does not understand it without V1, and it works either way with normal pleroma, so no reason to not do this
- Following HTTP Redirects when the HTTP Adapter is Finch
- Video framegrabs were not working correctly after the change to use Exile to execute ffmpeg
- Deactivated groups would still try to repeat a post.
- Fix logic error in Gun connection pooling which prevented retries even when the worker was launched with retry = true
- Connection pool errors when publishing an activity is a soft-error that will be retried shortly.
- Gun Connection Pool was not retrying to acquire a connection if the pool was full and stale connections were reclaimed
- TwitterAPI: Return proper error when healthcheck is disabled
- Handle cases when users.inbox is nil.
- Fix LDAP support
- Use correct domain for fqn and InstanceView
- The query for marking notifications as read has been simplified
- Mastodon API /api/v1/directory: Fix listing directory contents when not authenticated
- Ensure MediaProxy HTTP requests obey all the defined connection settings
- Fix a memory leak caused by Websocket connections that would not enter a state where a full garbage collection run could be triggered.
- Fix OpenGraph and Twitter metadata providers when parsing objects with no content or summary fields.
- MRF: Log sensible error for subdomains_regex
- MRF.StealEmojiPolicy: Properly add fallback extension to filenames missing one
- Federated timeline removal of hashtags via MRF HashtagPolicy
- Support objects with a null contentMap (firefish)
- Fix notifications query which was not using the index properly
- Notifications: improve performance by filtering on users table instead of activities table
- Prevent Rich Media backfill jobs from retrying in cases where it is likely they will fail again.
- Oban Jobs for refreshing users were not respecting the uniqueness setting
- Fix Optimistic Inbox for failed signatures
- MediaProxy Preview failures prevented when encountering certain video files
- pleroma_ctl: Use realpath(1) instead of readlink(1)
- ReceiverWorker: Make sure non-{:ok, _} is returned as {:error, …}
- Harden Rich Media parsing against very slow or malicious URLs
- Rich Media Preview cache eviction when the activity is updated.
- Parsing of RichMedia TTLs for Amazon URLs when query parameters are nil
- End of poll notifications were not streamed over websockets or web push
- Fix eblurhash and elixir-captcha not using system cflags
- Video thumbnails were not being generated due to a negative cache lookup logic error
- Fix web push notifications not successfully delivering
- Web Push notifications are no longer generated for muted/blocked threads and users.
- Fix validate_webfinger when running a different domain for Webfinger
### Removed
- Mastodon API: Remove deprecated GET /api/v1/statuses/:id/card endpoint https://github.com/mastodon/mastodon/pull/11213
- Removed support for multiple federator modules as we only support ActivityPub
MediaProxy was setting the content-length header which is not permitted by RFC9112§6.2 when we are chunking the reply as it conflicts with the existence of the transfer-encoding header.
Make remote emoji packs API use specifically the V1 URL. Akkoma does not understand it without V1, and it works either way with normal pleroma, so no reason to not do this
Remote object fetch failures will prevent the object fetch job from retrying if the object request returns 401, 403, 404, 410, or exceeds the maximum thread depth.
