When the evacuation of a nova-compute is in progress on a compute node,
starting the nova-compute service on that node would cause a race in
setting task_state [1] and could ultimately lead failures with all
future evacuation attempts on the compute node.
To avoid triggering that situation, introduce a startup wrapper when
Instance HA is enabled. The wrapper loops until the force-down flag is
clear and all evacuation finished, and then starts nova_compute.
[1] If71727cde51c29231dbb9a51c5babbcdfc802bdd
Partial-Bug: #1764883
Co-Authored-By: Andrew Beekhof <abeekhof@redhat.com>
Change-Id: Id1fc820b42fb72fc861fda82b04f6a3fa2b6b6f6
9602a9bafc