Ceph » RADOS

We'll keep this here in case we see it elsewhere, but the leaks I see are of messages and the AuthSessions associated with them.

So given that you're the first reporter, these are probably leaks coming out of the MDSMonitor portion of the code...

I took a look and didn't see anything going through MDSMonitor.* or FSCommand.*.

It looks like a leaked session during shutdown but I don't see how that happened nor a likely recent commit that caused it.

New one:

/ceph/teuthology-archive/yuriw-2018-01-23_20:26:59-multimds-wip-yuri-testing-2018-01-22-1653-luminous-testing-basic-smithi/2103508/remote/smithi202/log/valgrind/mon.a.log.gz

Still not seeing anything in RADOS runs AFAIK, but I did notice there might be some disparity in coverage....

13:33:31] <gregsfortytwo> batrick: also it still looks to me like we run the same valgrind config on fs and rados suites and those errors aren't popping up in rados runs
[13:34:08] <gregsfortytwo> ...although, hrm, the fs tests may stress the cluster more than our rados verify suite does
[13:34:10] <gregsfortytwo> interesting
[13:34:51] <gregsfortytwo> joshd: sage: so we only run valgrind against a mon_recovery, rados_api_tests, and rados_cls_all workloads
[13:35:24] <gregsfortytwo> there's also the singleton-flat runs but those are tagged expect_valgrind_errors so I'm not sure if they'd flag anything at us
[13:36:32] <gregsfortytwo> we thrash it a bit but nothing that would cause eg the client to reconnect
[13:37:28] <joshd> no ms error injection?
[13:38:12] <gregsfortytwo> oh, hrm, yeah msgr-failures/few.yaml is present

Patrick, which set of logs have the (Leak_DefinitelyLost, Leak_IndirectlyLost) errors?