Adar Dembo has posted comments on this change.
Change subject: KUDU-1090: relax MemTracker uniqueness constraint
Patch Set 2:
Line 128: // any crashes.
> might make more sense to put this in a different test case... this test in
Line 142: MonoDelta::FromMicroseconds(100));
> maybe you'd hit the race more with a super low value like 1us?
At first I tried 1us and 10 checker threads, but there was too much load and
cluster.Restart() timed out in 20% of runs. So I raised the poll period to 1ms,
and now 23/1000 runs fail with the MemTracker error, but I still can't reliably
repro it locally.
Do you think that's good enough?
Line 154: work.StopAndJoin();
> might be more fruitful if you set the log rolling threshold to be really lo
AFAICT the number of log segments is orthogonal to the number of times
tablet::RewindTabletForBootstrap is called (which is just once).
Line 156: // Restart the cluster. Nothing should crash.
> how about an AssertNoCrashes before the restart, too?
Line 237: LOG(FATAL) <<
> should we consider DFATAL here? would it actually cause incorrect behavior
It would lead to incorrect memory accounting, I think, but maybe that's the
worst of it.
I'll change to DFATAL and return the first found tracker.
To view, visit http://gerrit.cloudera.org:8080/4394
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings
Gerrit-Owner: Adar Dembo <a...@cloudera.com>
Gerrit-Reviewer: Adar Dembo <a...@cloudera.com>
Gerrit-Reviewer: Kudu Jenkins
Gerrit-Reviewer: Todd Lipcon <t...@apache.org>