Closed kostasrim closed 2 weeks ago
@adiholden @romange It looks like it's not a cluster issue. During shutdown we have a race condition, JournalSlice::AddLogRecord() execution can be preempted and the journal can be removed so when we proceed with execution the journal is already destroyed.
@BorysTheDev are you saying that UnregisterOnChange
has been called? but is not it under the cb_mu_
mutex?
@romange void ServerFamily::Shutdown() -> journal_->Close(); during JournalSlice::AddLogRecord() -> callback() because callback can preempt
It looks I was wrong regarding the reason, journal_->Close() doesn't destroy the JournalSlice
Currently is not reproducible and we decided with @adiholden to close it
https://github.com/dragonflydb/dragonfly/actions/runs/9938955158/job/27452529631#step:6:885
Trace: