Sync failing with 404 M_NOT_FOUND when the user is invited to a room with an unknown version

[t3chguy]

Description

As per the title; tried logging into my matrix.org test account and /sync responded with a 404 over and over again

Version information

• Homeserver: matrix.org

I believe the account was @webdevguru_test17:matrix.org in case that helps with log spelunking

It might be because the user was in an msc3083 unstable room the server dropped support for. 🔥

[babolivier]

This error is referring to an event sent in room !KQakPoXjBfiHdvvssg:matrix.org which indeed has the version org.matrix.msc3083. The event ($h0z7A514gZv7MmOt7y6Cck7Af2PcgDZDsmoEs8I5mC0) is @webdevguru_test22:matrix.org inviting @webdevguru_test17:matrix.org into the room.

[babolivier]

So what's happening here is:

• the sync handler loads the list of room memberships for this user
• it sees that the user is invited to the room
• it tries to get the invite event for this room using get_events, which eventually calls _get_events_from_db
• _get_events_from_db retrieves the event and checks the room version against known ones
• the room version isn't known so it ignores it
• _get_events_from_db returns no event, so get_events raises a 404 error

The check on the room version is here:

synapse/synapse/storage/databases/main/events_worker.py

Lines 781 to 801 in bdfde6d

	room_version = KNOWN_ROOM_VERSIONS.get(room_version_id)
	if not room_version:
	logger.warning(
	"Event %s in room %s has unknown room version %s",
	event_id,
	d["room_id"],
	room_version_id,
	)
	continue
	if room_version.event_format != format_version:
	logger.error(
	"Event %s in room %s with version %s has wrong format: "
	"expected %s, was %s",
	event_id,
	d["room_id"],
	room_version_id,
	room_version.event_format,
	format_version,
	)
	continue

[callahad]

Bumping to S-Critical as I'm assuming it blocks development work for @t3chguy (please let us know if it doesn't).

Bricking unstable rooms is one thing, but bricking accounts that have interacted with unstable rooms is quite another, and something I'm keen to see resolved prior to 1.40 going out.

[t3chguy]

It isn't blocking my development work, it is one of my ~30 test accounts, just happens to be the one my FF instance defaulted to and confused me why sync was returning 404.

Though if someone created such an unstable room (which Element Android release edition allowed you to do via UI) on their primary account then it'd be S-Critical for them

[clokep]

Just to define the impact I believe this will only cause an issue if you have an outstanding invite to an unknown room version, not users who have joined an unknown room version (see #7083).

Just to ensure we're all discussing the same thing:

• "unstable" is a room version that Synapse knows about but is not stable, according to the spec.
• "unknown" or "unsupported" is a room version which Synapse does not know about (and does not know how to interact with).
• "experimental" is generally a synonym for "unstable" but if the spec ever made a stable room version unstable than they would differ ("experimental" would then mean unspecced, e.g. from an MSC or something).