Ceph » CephFS

Ignore the last passing link above, that was actually a hammer run (which shows up as fs-master for some reason).

This was the last completely clean one: http://pulpito.ceph.com/teuthology-2017-01-21_17:15:02-fs-master---basic-smithi/ ( 11dce465f0f3ca53a38886b5ca169b17e74653f5)

And the last one without the characteristic four dead jobs was: http://pulpito.ceph.com/teuthology-2017-01-23_17:15:02-fs-master---basic-smithi/

So the range in which it got broken was:
20da91e881edee93666a3342f6178b25b1f4ca2f..d09fded5ae3a698219e4cf6f90b0d2e6dcbf7fbe

They're going dead during teardown when teuthology tries to list cephtest/ but something went wrong much earlier with the fuse mount (it didn't mount properly to begin with apparently).

After noticing http://tracker.ceph.com/issues/18838, I now also notice that the suspect commit range where this started failing does include the libcommon linkage change.

Working theory: something in the libcommon change is causing an issue that only manifests itself when linked with valgrind, that's causing ceph-fuse to get stuck somewhere during startup.