Lots of "Mismatched room_id for inbound group session" UTDs in threads

[robintown]

Steps to reproduce

Be in encrypted rooms where others are using threads heavily

Outcome

What did you expect?

No UTDs

What happened instead?

Threaded messages occasionally show up as "** Unable to decrypt: Error: Mismatched room_id for inbound group session (expected !ELvLSzljqBuAwjdBWx:t2l.io, was undefined) **"

Operating system

NixOS unstable

Browser information

Firefox 99.0.1

URL for webapp

develop.element.io

Application version

Element version: 9cb0185-react-e97536ef964b-js-6b5f4aa0a9c1 Olm version: 3.2.8

Homeserver

Synapse 1.58.0

Will you send logs?

Yes

[markiceman]

I have similar problem.

Element 1.10.12

Operating system: Windows

Browser: Firefox for Web version, Element desktop - both have the same problem

Homeserver: Synpase 1.58.1

Last message in thread in encrypted room is unreadable:

This is OK with Element 1.10.11 or Mobile App. It is problem with Element 1.10.12 only.

If I disable threads, message is readable.

Another screenshot with example:

[uhoreg]

It looks like Element is losing the room_id on some messages in threads, for some reason. If I "view source" on these undecryptable messages, then it shows me a blank Room ID

and expanding the "Original event source" shows no room_id field. But it obviously has a room ID, or at least had a room ID, somewhere along the line, since it got put in a room.

Is it possible that the event object gets passed to a function that removes the room_id field somehow?

[t3chguy]

This sounds very similar to the other threads issue which has thread.room is undefined/null

[flyingcakes85]

In case it matters, it just see this issue after I logged out and logged in.

[uhoreg]

It seems to happen mainly on the last message in a thread. Maybe there's a difference in backfilled messages vs messages from sync? Or something related to getting the megolm session before or after we get the message?

[Wendell250]

Happened to me today as well.

I wrote a message to a thread (on encrypted channel) and switched to another channel.
But when I came back the last message in the thread was replaced with this error.

Clearing cache and restarting Element didn't fix it.

Element desktop: 1.10.13

[ZReC]

It's happening to me every now and then.. only in threads. Sadly, I didn't find a consistent setup to trigger the glitch

[duxovni]

This isn't a real solution for the underlying issue, but as an interim fix while it's being investigated: the codepath throwing that error gets skipped if the provided room ID is null: https://github.com/matrix-org/matrix-js-sdk/blob/develop/src/crypto/OlmDevice.ts#L1066-L1073 Would it be reasonable to also skip it if the room ID is undefined?

[duxovni]

A tentative theory about what's actually going wrong:

• Synapse strips out the room_id field when serializing individual events in sync responses, because the events are already listed under their room ID in the sync response
• This serialization config is also applied to the latest thread reply event under m.relations
• Element Web re-inserts room IDs while processing the top-level events returned by a sync
• However, I don't see any sign of it re-inserting room IDs into events under m.relations. So that's a context where Element Web might ingest the latest_event of a thread with its room ID missing.