In-use retype fails if source deletion takes more than rpc timeout seconds

[Gorka Eguileor]

When doing a retype of an in-use volume (which requires migration) if the source volume deletion takes longer than rpc timeout seconds the migration will me marked as failed even though it will be successfully completed.

This is because Cinder Volume node calls Nova to do the copy and then Nova calls Cinder API for the completion and waits for the response, and Cinder API does a synchronous call to Cinder Volume and reports failure if there is a timeout on this RPC call.

Migration should not fail no matter how long the deletion takes, in the same way that we don't fail the migration just because deleting the source fails.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/211257

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/211257
Committed: https://git.openstack.org/cgit/openstack/cinder/commit/?id=cdf5e92f10fe8d097f91f6f63a2c77f6ab2bcdb8
Submitter: Jenkins
Branch: master

commit cdf5e92f10fe8d097f91f6f63a2c77f6ab2bcdb8
Author: Gorka Eguileor <email address hidden>
Date: Mon Aug 10 18:37:27 2015 +0200

    Make migration's volume source deletion async

    Currently source volume deletion on a volume migration (and therefore on
    retype with migration as well) is synchronous and that has 2 problems:

    1. If the deletion takes more than rpc_response_timeout then when
    performed on in-use volumes it will fail.

    2. If the deletion fails then the volume will be left in the back-end
    and Cinder will no longer have a reference of that volume in the DB.

    This patch makes source volume deletion asynchronous resolving both
    issues.

    Closes-Bug: #1483155
    Closes-Bug: #1483157
    Change-Id: I55c3d86660f90044a56b8609bb5774d4f5481227