Race Condition When Writing Filter to Disk

[naftulikay]

Race Condition When Writing Filter to Disk

[mtchavez]

Interesting. I will look into the issue. If you have any more information that could help debug that would be useful. One question I have is how large are your persisted filters on disk? Is it up to 4MiB?

[naftulikay]

They're around 4-10MiB.

Simply have a routine which writes the filter to disk in a for loop with a mutex, and observe the filesize.

[mtchavez]

Was definitely seeing some weird behavior. Want to try out the latest code in the master branch to see if this fixes the issue you are seeing before I tag a release?

[naftulikay]

Ah, I see. Looks good to me. I don't have access to the underlying code any more, but the solution makes sense. Since the file handles were closed either never or by the GC, there must have been a lot of stale data flowing undeterministically.

[mtchavez]

Great. Going to close this out then.