Bridge freezes after a minute of work

[JuniorJPDJ]

Hi, Today I updated bridge (I'm using master branch) and bridge freezes after a minute from restart. Last log message seems to be not related to the freeze. Last image i used and it was working correctly were 5 days old sha256:aad2827b27b23d31e5aa75fcd3f83cd1a9954762518b44c0a8ef461939249501, now I'm using image with tag latest (sha256:33f6fbd8f736167c2da51c9bf78628a17388a19750b62792b58054f66cfdeb38).

I'm unable to downgrade to last working version due to DB schema mismatch (my bad, I didn't backup it ;).

Some messages are getting backfilled from telegram chats, then bot freezes and doesn't send messages in any of the ways.

[JuniorJPDJ]

the same error persists in 0.11.3+dev.99344c38

[JuniorJPDJ]

The issue was with ffmpeg freezing whole bridge after freeze/long converting job. It should probably be handled another way to not block whole thread.

[Sharparam]

I seem to be experiencing this issue as well, on version 0.11.3+dev.e3a067c2 (via docker).

The last message to be printed in the log is:

[2022-08-20 14:24:53,265] [[email protected]] Starting direct file download in chunks of 65536 at 0, stride 65536

Before it freezes and will not process any further events until restarted, and then it freezes again with the same behaviour shortly after the restart, making the bridge unusable.

Looking at the process tree in htop shows an ffmpeg process under mautrix-telegram that shows no sign of doing any actual work.

[chagai95]

I think I'm facing a similar issue, did you manage to solve this? I don't see how ffmpeg in htop is related to mautrix- telegram but it sounds like the exact problem I'm having...I tried looking the process but it just got started again...I also can't downgrade because I made no recent backup but I think I have a backup from March, maybe I'll try that if there's no other solution...

[chagai95]

I disabled backfill and it seems to work fine again 😍

[verymilan]

looks like this is the issue i am having as well.. i hope setting the missed_limit to 0 really is all it takes to disable it. i at least can confirm that killing the ffmpeg process brings the bridge back to life.

[verymilan]

update: disabling backfill only delayed the problem from recurring

[chagai95]

So you have the issue again? I haven't had it for a long time

[verymilan]

yes, worked right until my second comment ... and so far until now after killing ffmpeg

[chagai95]

Ok thx, hope it doesn't come back