UTDs not retried when key arrives at almost the same time

[andybalaam]

Steps to reproduce

See this rageshake: https://github.com/element-hq/element-x-android-rageshakes/issues/4076

The logs in the receiver side show a pattern like this:

2025-01-20T15:46:43.519041Z Failed to decrypt a room event
... (9 more Failed to decrypt)
2025-01-20T15:46:46.398276Z Failed to decrypt a room event
2025-01-20T15:46:46.407885Z Received a new megolm room key
2025-01-20T15:46:46.418528Z Failed to decrypt a room event
.. (2 more Failed to decrypt)

So the megolm key arrived at almost the same moment as the events.

There is no mention of "Retrying decryption" anywhere in the receiver logs.

I am wondering whether the async task pulling from room_keys_received_stream() might have retried decryption too early, before the incoming batch of events was finished processing?

[andybalaam]

Could be caused by event cache work?

[andybalaam]

https://github.com/element-hq/element-android-rageshakes/issues/75459 could be another instance

[richvdh]

I think matrix-org/matrix-rust-sdk#4715 rewrote a bunch of this logic, but unfortunately the retry task's logging is at debug, which doesn't seem to appear in the EXA logs.

[richvdh]

Seems like this affects keys from backup as well as keys received from other devices

[richvdh]

Hopefully this will be fixed by matrix-org/matrix-rust-sdk#3872

[andybalaam]

Another example: https://github.com/element-hq/element-x-android-rageshakes/issues/5174