We’ve received reports of some specific instances slowly accumulating
more and more binary data over time up to OOMs and globally setting
ERL_FULLSWEEP_AFTER=0 has proven to be an effective countermeasure.
However, this incurs increased cpu perf costs everywhere and is
thus not suitable to apply out of the box.
Apparently long-lived Phoenix websocket processes are known to
often cause exactly this by getting into a state unfavourable
for the garbage collector.
Therefore it seems likely affected instances are using timeline
streaming and do so in just the right way to trigger this. We
can tune the garbage collector just for websocket processes
and use a more lenient value of 20 to keep the added perf cost
in check.
Testing on one affected instance appears to confirm this theory
Ref.:
https://www.erlang.org/doc/man/erlang#ghlink-process_flag-2-idp226https://blog.guzman.codes/using-phoenix-channels-high-memory-usage-save-money-with-erlfullsweepafterhttps://git.pleroma.social/pleroma/pleroma/-/merge_requests/4060
Tested-by: bjo
Websites are increasingly getting more bloated with tricks like inlining content (e.g., CNN.com) which puts pages at or above 5MB. This value may still be too low.
Rich Media parsing was previously handled on-demand with a 2 second HTTP request timeout and retained only in Cachex. Every time a Pleroma instance is restarted it will have to request and parse the data for each status with a URL detected. When fetching a batch of statuses they were processed in parallel to attempt to keep the maximum latency at 2 seconds, but often resulted in a timeline appearing to hang during loading due to a URL that could not be successfully reached. URLs which had images links that expire (Amazon AWS) were parsed and inserted with a TTL to ensure the image link would not break.
Rich Media data is now cached in the database and fetched asynchronously. Cachex is used as a read-through cache. When the data becomes available we stream an update to the clients. If the result is returned quickly the experience is almost seamless. Activities were already processed for their Rich Media data during ingestion to warm the cache, so users should not normally encounter the asynchronous loading of the Rich Media data.
Implementation notes:
- The async worker is a Task with a globally unique process name to prevent duplicate processing of the same URL
- The Task will attempt to fetch the data 3 times with increasing sleep time between attempts
- The HTTP request obeys the default HTTP request timeout value instead of 2 seconds
- URLs that cannot be successfully parsed due to an unexpected error receives a negative cache entry for 15 minutes
- URLs that fail with an expected error will receive a negative cache with no TTL
- Activities that have no detected URLs insert a nil value in the Cachex :scrubber_cache so we do not repeat parsing the object content with Floki every time the activity is rendered
- Expiring image URLs are handled with an Oban job
- There is no automatic cleanup of the Rich Media data in the database, but it is safe to delete at any time
- The post draft/preview feature makes the URL processing synchronous so the rendered post preview will have an accurate rendering
Overall performance of timelines and creating new posts which contain URLs is greatly improved.
Trying to display non-media as media crashed the renderer,
but when posting a status with a valid, non-media object id
the post was still created, but then crashed e.g. timeline rendering.
It also crashed C2S inbox reads, so this could not be used to leak
private posts.
In Mastodon media can only be used by owners and only be associated with
a single post. We currently allow media to be associated with several
posts and until now did not limit their usage in posts to media owners.
However, media update and GET lookup was already limited to owners.
(In accordance with allowing media reuse, we also still allow GET
lookups of media already used in a post unlike Mastodon)
Allowing reuse isn’t problematic per se, but allowing use by non-owners
can be problematic if media ids of private-scoped posts can be guessed
since creating a new post with this media id will reveal the uploaded
file content and alt text.
Given media ids are currently just part of a sequentieal series shared
with some other objects, guessing media ids is with some persistence
indeed feasible.
E.g. sampline some public media ids from a real-world
instance with 112 total and 61 monthly-active users:
17.465.096 at t0
17.472.673 at t1 = t0 + 4h
17.473.248 at t2 = t1 + 20min
This gives about 30 new ids per minute of which most won't be
local media but remote and local posts, poll answers etc.
Assuming the default ratelimit of 15 post actions per 10s, scraping all
media for the 4h interval takes about 84 minutes and scraping the 20min
range mere 6.3 minutes. (Until the preceding commit, post updates were
not rate limited at all, allowing even faster scraping.)
If an attacker can infer (e.g. via reply to a follower-only post not
accessbile to the attacker) some sensitive information was uploaded
during a specific time interval and has some pointers regarding the
nature of the information, identifying the specific upload out of all
scraped media for this timerange is not impossible.
Thus restrict media usage to owners.
Checking ownership just in ActivitDraft would already be sufficient,
since when a scheduled status actually gets posted it goes through
ActivityDraft again, but would erroneously return a success status
when scheduling an illegal post.
Independently discovered and fixed by mint in Pleroma
1afde067b1
In MastoAPI media descriptions are updated via the
media update API not upon post creation or post update.
This functionality was originally added about 6 years ago in
ba93396649 which was part of
https://git.pleroma.social/pleroma/pleroma/-/merge_requests/626 and
https://git.pleroma.social/pleroma/pleroma-fe/-/merge_requests/450.
They introduced image descriptions to the front- and backend,
but predate adoption of Mastodon API.
For a while adding an `descriptions` array on post creation might have
continued to work as an undocumented Pleroma extension to Masto API, but
at latest when OpenAPI specs were added for those endpoints four years
ago in 7803a85d2c, these codepaths ceased
to be used. The API specs don’t list a `descriptions` parameter and
any unknown parameters are stripped out.
The attachments_from_ids function is only called from
ScheduledActivity and ActivityDraft.create with the latter
only being called by CommonAPI.{post,update} whihc in turn
are only called from ScheduledActivity again, MastoAPI controller
and without any attachment or description parameter WelcomeMessage.
Therefore no codepath can contain a descriptions parameter.
Due to JSON-LD compaction the full address of public scope
may also occur in shorter forms and the spec requires us to treat them
all equivalently. To save us the pain of repeatedly checking for all
variants internally, normalise inbound data to just one form.
See note at: https://www.w3.org/TR/activitypub/#public-addressing
This needs to happen very early, even before the other addressing fixes
else an earlier validator will reject the object. This in turn required
to move the list-tpye normalisation earlier as well, but since I was
unsure about putting empty lists into the data when no such field
existed before, I excluded this case and thus the later fixing had to be
kept as well.
Fixes: #670
