Closed lfcnassif closed 10 months ago
Sure, I will take a look at this today.
I finished my changes and tests, this looks fine.
Just an observation for others: audios that might theoretically cause crashes in python transcription processes were skipped intentionally, to avoid infinite loops on client side trying to transcribe the audio on different nodes. But as I think this is very unlikely to happen (corrupted audios often fail or trigger a timeout in the wav conversion step), and since we have this current situation of python processes crashing because of kernel or driver updates, I pushed commit 653dca1 to retry audios if python crashes, so clients won't skip audios silently. If some client hits some infinite loop with some corrupted audio, it won't be a silent issue, the client will complain and we can take a closer look at his audio to try to handle it properly.
@hauck-jvsh tested this on a Kubernetes test environment and it worked, so I'm merging. Thank you @hauck-jvsh!
Fix #1942.
@hauck-jvsh, I'm still going to do tests, but I would appreciate your review, thanks.