Ceph » CephFS

The MultiRankRecovery test has some unstable checks at the end. The test is built to have a quiesce cluster reach QUIESCED state on a set and then the cluster is subjected to a series of membership updates.

At one point, a quiescing peer is added back to the cluster. The system is expected to query the new peer and immediately know that it's quiesced, thus preventing a transition from QUIESCED to QUIESCING on the said set.

Next, a non-quiescing peer is added, and a set is expected to move to QUIESCING.

https://jenkins.ceph.com/job/ceph-pull-requests-arm64/56600/consoleFull#-355712461e840cee4-f4a4-4183-81dd-42855615f2c1 is an example of a failure of this test.

/home/jenkins-build/build/workspace/ceph-pull-requests-arm64/src/test/mds/TestQuiesceDb.cc:1607: Failure
Expected equality of these values:
  QS_QUIESCED
  last_request->response.sets.at("set1").rstate.state
    Which is: QS_QUIESCING
/home/jenkins-build/build/workspace/ceph-pull-requests-arm64/src/test/mds/TestQuiesceDb.cc:1608: Failure
Expected equality of these values:
  QS_QUIESCED
  last_request->response.sets.at("set2").rstate.state
    Which is: QS_QUIESCING