Ceph

Workaround¶

Run ceph-disk activate manually.

Rationale for Won't fix¶

The problem revealed by this suite is a race condition that can be fixed easily manually. It is fixed in infernalis, together with a number of fixes related to udev related fixes. It would not be trivial to backport on hammer and it looks like people usually work around this.

Description¶

curl --silent http://paddles.front.sepia.ceph.com/runs/?suite=ceph-deploy | jq '.[] | .name' | while read run ; do eval run=$run ; curl --silent http://paddles.front.sepia.ceph.com/runs/$run/jobs/ | jq '.[] | select(.os_version == "6.5" and .status == "pass") | select(.description | contains("dmcrypt")) | .name' ; done 

returns nothing and

teuthology-suite --verbose --suite ceph-deploy --filter="ceph-deploy/basic/{ceph-deploy-overrides/ceph_deploy_dmcrypt.yaml config_options/cephdeploy_conf.yaml distros/centos_6.5.yaml tasks/ceph-deploy_hello_world.yaml}" --suite-branch wip-ceph-deploy-test-hammer --ceph hammer-backports --machine-type vps --priority 101

fails consistently (see http://pulpito.ceph.com/loic-2015-10-05_01:40:39-ceph-deploy-hammer-backports---basic-vps/1088457/ for an example).

I suspect this is a combination of incorrect udev rules and racing udev events generated by ceph-disk.

teuthology-suite --verbose --suite ceph-deploy --filter="ceph-deploy/basic/{ceph-deploy-overrides/ceph_deploy_dmcrypt.yaml config_options/cephdeploy_conf.yaml distros/centos_6.5.yaml tasks/ceph-deploy_hello_world.yaml}" --suite-branch wip-ceph-deploy-test-hammer --ceph hammer-backports --machine-type vps --priority 101

will create a teuthology job that never returns (it waits forever for the cluster to be healthy) which is convenient to investigate.