Recover from Build state on compute manager start-up

[Phil Day]

If a compute manager is stopped / fails during a build operation then the instance will be left stuck with vm_state=BUILDING

During restart the one thing we can be sure about such instances is that providing the task state is not SCHEDULING (i.e the request is not still on the queue) then there is no thread running for this instance. (Might need to add another task_state to be set as soon as possible on the the compute manager).

In this case it should be possible to treat instances in this state as if they have failed to spawn and either put them into an ERROR state, or even better tidy-up and send back to the scheduler as if the spawn had failed.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/47836

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/47836
Committed: http://github.com/openstack/nova/commit/0e3d2622ae7e1fce85cda1e7451bcc932d116fbf
Submitter: Jenkins
Branch: master

commit 0e3d2622ae7e1fce85cda1e7451bcc932d116fbf
Author: David McNally <email address hidden>
Date: Mon Sep 23 14:31:15 2013 +0100

    Recover from build state on compute manager start-up

    If a compute manager is stopped / fails during a build operation
    then the instance will be left stuck with vm_state=BUILDING

    During restart the one thing we can be sure about such instances
    is that providing the task state is not SCHEDULING (i.e the request
    is not still on the queue) then there is no thread running for this
    instance.

    In this case we can treat instances in this state as if they have
    failed to spawn and put the into an ERROR state.

    Closes-Bug: 1197024
    Related to blueprint recover-stuck-state

    Change-Id: I7b116d8154036c43b0feb23f5669f52358408d2b