Crash after stopping instance about dragonfly HOT 6 CLOSED

I was hoping that it would fix that, but it looks like it doesn't.
I created a PR in helio to fix the traversal code, that's likely to be related to the ctrl-c issue. I guess the malloc_trim is unrelated. Bummer.

from dragonfly.

This seems to be a dup of #2714
Unfortunately, it doesn't reproduce for me locally, even with --count=1000 😱
However, the call stack here helps. I imagine that it might be that we're removing the listener from conn_list and then call Listener::PreShutdown() which calls DispatchTracker::TrackAll(), which eventually calls ListenerInterface::TraverseConnectionsOnThread. I'm not entirely sure why this can happen, but we can easily add if (it == end()) return; here:
https://github.com/romange/helio/blob/8985263c3acca038752e8f9fdd8e9f61d2ec2b6f/util/fibers/listener_interface.cc#L282

from dragonfly.

@chakaz @adiholden also reproduces on our sanitizers job (similar error different context) https://github.com/dragonflydb/dragonfly/actions/runs/8337372698/job/22816007105#step:8:2880

it fails because of malloc_trim

from dragonfly.

@kostasrim @adiholden can you please help me and see if you can reproduce it in main updated with #2742 ? 🙏

from dragonfly.

@chakaz is it suppose to fix malloc_trim as well? Sanitizers failed (and I already run them yesterday against your branch). I can try with ctr-c instead (maybe they were separate issues 🤷‍♂️ )

from dragonfly.

@chakaz I will open a separate issue then

from dragonfly.