On 3/19/25 14:43, Tomas Vondra wrote:
>
> ...
>
> I'll get this fixed shortly. Unfortunately, this means the "bloom"
> filters may be broken - not just those built in parallel, the union
> method can be triggered due to concurrent activity too.
>
Here's a more complete version of the fix, along with a proper commit
message, etc.
While working on this, I realized there's a second (less severe issue,
in that it fails to free the decompressed filters. I believe this would
be mostly harmless before parallel builds, because we'd merge only one
summary at a time, I think. With parallel builds it may be much worse,
but I'm yet to try how bad.
FWIW I think the minmax-multi has a similar memory leak.
regards
--
Tomas Vondra