Implement synapse issue #16751: Treat local_media_directory as optional storage provider#19204
Conversation
… as optional storage provider This allows off-site media storage without local cache by making the local media directory behave like any other MediaStorageProvider. When no local directory is configured, media is stored directly to remote providers only, with temporary files used for thumbnail generation when needed.
… as optional storage provider This allows off-site media storage without local cache by making the local media directory behave like any other MediaStorageProvider. When no local directory is configured, media is stored directly to remote providers only, with temporary files used for thumbnail generation when needed. Also fixes URL cache thumbnails to be served directly from local filesystem instead of being incorrectly routed through storage providers.
These were auto-formatted changes that weren't intended to be committed.
Add type ignore comment for os.path.join call since local_provider check guarantees local_media_directory is not None.
anoadragon453
left a comment
There was a problem hiding this comment.
Thank you for your contribution. Several questions below.
From the linked issues, I think broadly what we're aiming for is to satisfy the use-case where media isn't stored on disk at all, and is only stored on (typically) remote providers.
"nit" below means a non-blocking change. Typically related to code style or readability.
Address code review feedback: - Add enable_local_media_storage config option to disable local media store - Refactor MediaStorage.__init__ to accept local_provider as optional FileStorageProviderBackend argument - Rename fname to media_filepath and eliminate dirname variable - Deduplicate spam check code in store_into_file - Delete temp files immediately after upload to storage providers - Fix fetch_media to check local_provider first - Remove unnecessary list() wrapper - Revert storage_providers type to Sequence[StorageProvider] - Update tests to use new MediaStorage constructor signature
When no local provider is configured, temp files are now automatically cleaned up after use. This addresses the reviewer feedback about temp files building up on disk over time. The method now yields the file path within a context manager, and callers must use `async with` to ensure proper cleanup.
|
@anoadragon453 thanks for the review. I reworked the suggested parts and found another area (the thumbnail generation) that needed fixing. Hope this is better? 😄 |
Add enable_local_media_storage to the JSON schema and regenerate the configuration documentation so the new flag appears under the Media Store section.
There was a problem hiding this comment.
Looking mostly good now! Thanks for making those changes.
Couple more things, plus the following unit tests are currently failing:
tests.replication.test_multi_media_repo.AuthenticatedMediaRepoShardTestCase.test_download_simple_file_racetests.replication.test_multi_media_repo.MediaRepoShardTestCase.test_download_simple_file_race
These tests assume that the files we fetch remain on disk, and this assumption should continue to hold as enable_local_media_storage is True in tests.
It'd also be good to see some tests with enable_local_media_storage set to False, so we can check the happy path works, and that files aren't pooling up on disk.
Note that you can use the @override_config decorator on tests to dynamically modify homeserver configuration for just that test case.
- Move database insert outside store_into_file context manager to ensure proper cleanup of duplicate files on constraint violations - Add explicit file cleanup on IntegrityError when concurrent downloads create duplicate files - Refactor duplicate exception handling into _store_remote_media_with_cleanup helper method - Add tests for enable_local_media_storage=False to verify no local file accumulation when local storage is disabled - Remove unnecessary ensure_media_is_in_local_cache call in thumbnailer.py Fixes race condition where two workers concurrently downloading the same media would result in two files being stored instead of one.
|
I think I fixed all tests and also added the suggested tests. It's a bit tricky to get proper tests with external providers up and running so I went with a compromise. |
anoadragon453
left a comment
There was a problem hiding this comment.
Just a few small changes, and I think this is good to go! Thanks for your patience with the back and forth.
Co-authored-by: Andrew Morgan <[email protected]>
Co-authored-by: Andrew Morgan <[email protected]>
When enable_local_media_storage is set to false and no media_storage_providers are configured, all media requests will return 404 errors as there will be no storage backend available. This addresses reviewer feedback to document the expected behavior in this configuration scenario.
When enable_local_media_storage is disabled and no media_storage_providers are configured, log a warning at startup to inform administrators that all media requests will return 404 errors due to the lack of storage backend.
anoadragon453
left a comment
There was a problem hiding this comment.
Excited to see this land. Thanks for your patience with the review process!
Yay. No worries. We have to get this right! |
Ref: - #4882 - element-hq/synapse#19204 - https://element-hq.github.io/synapse/latest/usage/configuration/config_documentation.html#enable_local_media_storage We default it to `true`, keeping up with upstream and the old behavior. s3-storage-provider users may set `matrix_synapse_enable_local_media_storage` to `false` to disable local file caching. This likely comes at the expense of some performance. For matrix-media-repo users, it likely doesn't matter what this is set to, as for a matrix-media-repo setup, all media-related API endpoints are captured and forwarded to matrix-media-repo (before reaching Synapse).
This PR contains the following updates: | Package | Update | Change | |---|---|---| | [element-hq/synapse](https://github.com/element-hq/synapse) | minor | `1.145.0` → `1.146.0` | --- >⚠️ **Warning** > > Some dependencies could not be looked up. Check the Dependency Dashboard for more information. --- ### Release Notes <details> <summary>element-hq/synapse (element-hq/synapse)</summary> ### [`v1.146.0`](https://github.com/element-hq/synapse/releases/tag/v1.146.0) [Compare Source](element-hq/synapse@v1.145.0...v1.146.0rc1) ### Synapse 1.146.0 (2026-01-27) No significant changes since 1.146.0rc1. #### Deprecations and Removals - [MSC2697](matrix-org/matrix-spec-proposals#2697) (Dehydrated devices) has been removed, as the MSC is closed. Developers should migrate to [MSC3814](matrix-org/matrix-spec-proposals#3814). ([#​19346](element-hq/synapse#19346)) - Support for Ubuntu 25.04 (Plucky Puffin) has been dropped. Synapse no longer builds debian packages for Ubuntu 25.04. ### Synapse 1.146.0rc1 (2026-01-20) #### Features - Add a new config option [`enable_local_media_storage`](https://element-hq.github.io/synapse/latest/usage/configuration/config_documentation.html#enable_local_media_storage) which controls whether media is additionally stored locally when using configured `media_storage_providers`. Setting this to `false` allows off-site media storage without a local cache. Contributed by Patrice Brend'amour [@​dr](https://github.com/dr).allgood. ([#​19204](element-hq/synapse#19204)) - Stabilise support for [MSC4312](matrix-org/matrix-spec-proposals#4312 `m.oauth` User-Interactive Auth stage for resetting cross-signing identity with the OAuth 2.0 API. The old, unstable name (`org.matrix.cross_signing_reset`) is now deprecated and will be removed in a future release. ([#​19273](element-hq/synapse#19273)) - Refactor Grafana dashboard to use `server_name` label (instead of `instance`). ([#​19337](element-hq/synapse#19337)) #### Bugfixes - Fix joining a restricted v12 room locally when no local room creator is present but local users with sufficient power levels are. Contributed by [@​nexy7574](https://github.com/nexy7574). ([#​19321](element-hq/synapse#19321)) - Fixed parallel calls to `/_matrix/media/v1/create` being ratelimited for appservices even if `rate_limited: false` was set in the registration. Contributed by [@​tulir](https://github.com/tulir) @​ Beeper. ([#​19335](element-hq/synapse#19335)) - Fix a bug introduced in 1.61.0 where a user's membership in a room was accidentally ignored when considering access to historical state events in rooms with the "shared" history visibility. Contributed by Lukas Tautz. ([#​19353](element-hq/synapse#19353)) - [MSC4140](matrix-org/matrix-spec-proposals#4140): Store the JSON content of scheduled delayed events as text instead of a byte array. This fixes the inability to schedule a delayed event with non-ASCII characters in its content. ([#​19360](element-hq/synapse#19360)) - Always rollback database transactions when retrying (avoid orphaned connections). ([#​19372](element-hq/synapse#19372)) - Fix `InFlightGauge` typing to allow upgrading to `prometheus_client` 0.24. ([#​19379](element-hq/synapse#19379)) #### Updates to the Docker image - Add [Prometheus HTTP service discovery](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#http_sd_config) endpoint for easy discovery of all workers when using the `docker/Dockerfile-workers` image (see the [*Metrics* section of our Docker testing docs](docker/README-testing.md#metrics)). ([#​19336](element-hq/synapse#19336)) #### Improved Documentation - Remove docs on legacy metric names (no longer in the codebase since 2022-12-06). ([#​19341](element-hq/synapse#19341)) - Clarify how the estimated value of room complexity is calculated internally. ([#​19384](element-hq/synapse#19384)) #### Internal Changes - Add an internal `cancel_task` API to the task scheduler. ([#​19310](element-hq/synapse#19310)) - Tweak docstrings and signatures of `auth_types_for_event` and `get_catchup_room_event_ids`. ([#​19320](element-hq/synapse#19320)) - Replace usage of deprecated `assertEquals` with `assertEqual` in unit test code. ([#​19345](element-hq/synapse#19345)) - Drop support for Ubuntu 25.04 'Plucky Puffin', add support for Ubuntu 25.10 'Questing Quokka'. ([#​19348](element-hq/synapse#19348)) - Revert "Add an Admin API endpoint for listing quarantined media ([#​19268](element-hq/synapse#19268))". ([#​19351](element-hq/synapse#19351)) - Bump `mdbook` from 0.4.17 to 0.5.2 and remove our custom table-of-contents plugin in favour of the new default functionality. ([#​19356](element-hq/synapse#19356)) - Replace deprecated usage of PyGitHub's `GitRelease.title` with `.name` in release script. ([#​19358](element-hq/synapse#19358)) - Update the Element logo in Synapse's README to be an absolute URL, allowing it to render on other sites (such as PyPI). ([#​19368](element-hq/synapse#19368)) - Apply minor tweaks to v1.145.0 changelog. ([#​19376](element-hq/synapse#19376)) - Update Grafana dashboard syntax to use the latest from importing/exporting with Grafana 12.3.1. ([#​19381](element-hq/synapse#19381)) - Warn about skipping reactor metrics when using unknown reactor type. ([#​19383](element-hq/synapse#19383)) - Add support for reactor metrics with the `ProxiedReactor` used in worker Complement tests. ([#​19385](element-hq/synapse#19385)) </details> --- ### Configuration 📅 **Schedule**: Branch creation - At any time (no schedule defined), Automerge - At any time (no schedule defined). 🚦 **Automerge**: Disabled by config. Please merge this manually once you are satisfied. ♻ **Rebasing**: Whenever PR is behind base branch, or you tick the rebase/retry checkbox. 🔕 **Ignore**: Close this PR and you won't be reminded about this update again. --- - [ ] <!-- rebase-check -->If you want to rebase/retry this PR, check this box --- This PR has been generated by [Renovate Bot](https://github.com/renovatebot/renovate). <!--renovate-debug:eyJjcmVhdGVkSW5WZXIiOiI0Mi42OS4yIiwidXBkYXRlZEluVmVyIjoiNDIuNjkuMiIsInRhcmdldEJyYW5jaCI6Im1haW4iLCJsYWJlbHMiOlsiaW1hZ2UiXX0=--> Reviewed-on: https://gitea.alexlebens.dev/alexlebens/infrastructure/pulls/3533 Co-authored-by: Renovate Bot <[email protected]> Co-committed-by: Renovate Bot <[email protected]>
# Famedly Synapse Release v1.146.0_1 depends on: famedly/complement#10 ## Famedly additions for v1.146.0_1 - feat: trigger CI actions (that are triggered on PRs) in merge queue (FrenchGithubUser) ### Notes for Famedly: #### Deprecations and Removals - matrix-org/matrix-spec-proposals#2697 (Dehydrated devices) has been removed, as the MSC is closed. Developers should migrate to matrix-org/matrix-spec-proposals#3814. (element-hq/synapse#19346) - Support for Ubuntu 25.04 (Plucky Puffin) has been dropped. Synapse no longer builds debian packages for Ubuntu 25.04. #### Updates to the Docker image - Add [Prometheus HTTP service discovery](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#http_sd_config) endpoint for easy discovery of all workers when using the docker/Dockerfile-workers image (see the [Metrics section of our Docker testing docs](https://github.com/famedly/synapse/pull/docker/README-testing.md#metrics)). (element-hq/synapse#19336) #### Features - Add a new config option [enable_local_media_storage](https://element-hq.github.io/synapse/latest/usage/configuration/config_documentation.html#enable_local_media_storage) which controls whether media is additionally stored locally when using configured media_storage_providers. Setting this to false allows off-site media storage without a local cache. Contributed by Patrice Brend'amour @dr.allgood. (element-hq/synapse#19204) - Stabilise support for matrix-org/matrix-spec-proposals#4312 m.oauth User-Interactive Auth stage for resetting cross-signing identity with the OAuth 2.0 API. The old, unstable name (org.matrix.cross_signing_reset) is now deprecated and will be removed in a future release. (element-hq/synapse#19273) - Refactor Grafana dashboard to use server_name label (instead of instance). (element-hq/synapse#19337)
This allows off-site media storage without local cache by making the local media directory behave like any other MediaStorageProvider. When no local directory is configured, media is stored directly to remote providers only, with temporary files used for thumbnail generation when needed.
fixes #16751
Pull Request Checklist
EventStoretoEventWorkerStore.".code blocks.