ceph redeployment should not use an old clusters fetch directory

[John Fulton]

If the LocalCephAnsibleFetchDirectoryBackup [1] contains a backup for the ceph-ansible fetch directory for a ceph cluster from a previous deployment (with a different fsid), then do not use that fetch directory on a new deployment.

[1] https://github.com/openstack/tripleo-heat-templates/blob/master/deployment/ceph-ansible/ceph-base.yaml#L177

[John Fulton (jfulton-org)]

Make sure you delete your overcloud correctly when testing the proposed fix [0] or you'll get the same FSID for each deployment.

If you just delete your overcloud with Heat like this, then the deployment plan is not deleted and the old FSID will be stored there. Then when you test the proposed change it won't work because you'll get the same FSID each time you deploy what should be a new overcloud.

 openstack stack delete $overcloud_name

This is because the FSID is generated by the tripleo client [1] and then stored in the deployment plan. The deployment plan is deleted when you correctly delete your overcloud using a command like this:

 openstack overcloud delete $overcloud_name

[0] https://review.openstack.org/#/c/652062
[1] https://github.com/openstack/tripleo-common/blob/master/tripleo_common/utils/passwords.py#L56

[OpenStack Infra (hudson-openstack)]

Change abandoned by John Fulton (<email address hidden>) on branch: master
Review: https://review.opendev.org/652062
Reason: The fetch directory contains sub directories named after the FSID. Only the directory named after the FSID will be used during the deployment the others will be ignored. Though this could save a little time by skipping tasks to unpack the fetch directory it could also add extra complication and shouldn't be necessary.

[John Fulton (jfulton-org)]

Ceph redeployment will not use an old cluster's fetch directory because the fetch directory is only reused if the FSID matches. This is already the case because the items in fetch directory are only accessed using the path /path/to/fetch/$FSID/$item. Because the FSID is unique per deployment of Ceph, /path/to/fetch/$FSID/$item will always refer to the fetch directory appropriate for the deployment.

If the overcloud is not correctly deleted and the deployment plan remains because someone only deleted using Heat, "openstack stack delete overcloud", then that's a mistake. The correct way to delete the overcloud is "openstack overcloud delete overcloud". This will also remove the deployment plan so that the FSID in the deployment plan is not reused. Ideally we won't depend on the fetch directory but it shouldn't' harm anything provided that TripleO is used correctly.