Snapshot creation is racey

[Rick Harris]

A race condition exists in the creation of snapshots since we check the task_state in compute/api but don't set the task_state until compute/manager receives the message.

This can allow two (or more) snapshots to be created at virtually the same time, which, in the case of XenServer may end up causing VHDs to not coalesce properly.

Even worse, when the VHD doesn't coalesce, this can cause the creation of "bad" images which ultimately corrupt the SR on the machine which they are restored to. This can have a cascading effect across the cluster where one bad image corrupts multiple machines.

The two pronged solution is to prevent "bad" images from getting created in the first place, and, 2), if we detect a bad image, bail on it before corrupting the SR.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/5059

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/5059
Committed: http://github.com/openstack/nova/commit/08b4e6c2b808011ea7ae9b367bfb829cb332f4e7
Submitter: Jenkins
Branch: master

commit 08b4e6c2b808011ea7ae9b367bfb829cb332f4e7
Author: Rick Harris <email address hidden>
Date: Thu Mar 8 02:55:04 2012 +0000

    Fix racey snapshots.

    Fixes bug 949475

    Atomically tests and sets the instance task_state before allowing a
    snapshot or backup to be initiated.

    Change-Id: I40671a80f5e75337e176a715837f62d400cc21b6