Brian Anderson
Brian Anderson
I moved this out of draft because these patches are complete and can be merged as is if desired, but I haven't completed the writeup. I'd like to throw a...
I pushed a revert temporarily to get the CI to show the crash.
Here's the crash under CI. This is interesting because it is not how I see the crash manifest locally... (this says "reached unreachable code", but the one i see is...
Status on this is that I have some confidence the fix does fix the eviction crash, but not the windows crash; the test case I added is quite invasive and...
> Here are two high-level considerations: > > * Eviction and Shutdown are very similar, and I feel we should try to share the synchronization mechanism already present in `client.close()`...
I have rebased and pushed just the fix with no tests. This can be merged as-is if desired. But there is no regression test. The big problem right now is...
> > I can't find a `close` method on client or context and don't see what synchonization mechanism this refers to. > > Sorry Brian, I was thinking of the...
Current test failure doesn't look related.
I also want to point out one other problem with the current code that I didn't write in the op: because the current code calls (rarely) a user-defined callback on...
Current ci failure seems to be real: https://github.com/tigerbeetle/tigerbeetle/actions/runs/19877261564/job/56968547845?pr=3373 I have to sleep on this. I am so frustrated.