src/osd/PG.cc: 6455: FAILED assert(0 == "we got a bad state machine event")

[dzafman]

No description provided.

[dzafman]

I ran a failing test run 20 times it passed every time. Not sure how hard it was to hit this:

http://pulpito.ceph.com/dzafman-2018-03-16_20:00:06-rados-wip-zafman-testing2-distro-basic-smithi/

[dzafman]

Should we run a subset of rados/thrash suite?

[xiexingguo]

I ran a failing test run 20 times it passed every time. Not sure how hard it was to hit this

@dzafman It might not be as rare as it is thought to be. I've hit the same issue a couple of weeks ago and posted a cure at #20837. It is actual a replica-backfill-revoke vs primary-backfill-finish race. Well, I'd confess this is a better fix:-)

[neha-ojha]

@dzafman It appeared quite frequently for me when I ran rados runs with --subset 3/500. But yes, it was difficult to reproduce in a single test.

[jdurgin]

Looks good to me, should be on the lookout for this reoccurring in future runs. Running it through a rados subset like neha was doing seems good enough for now

[dzafman]

http://qa-proxy.ceph.com/teuthology/dzafman-2018-03-21_09:57:19-rados:thrash-wip-zafman-testing2-distro-basic-smithi/

2312022 DEAD Infrastructure
2312125 FAIL scrub stat mismatch
2312148 DEAD Thrasher kept going even after no more client activity

[gregsfortytwo]

LGTM