Get instances from conductor in init_host.

Update compute's init_host() to get the list of instances on this host
via the conductor service.  This removes a db read from compute.

Most of the test fixes are to ensure conductor is running anywhere that
compute is running.  The EC2 availability zones test change (from 13 to
15) is because this output includes info on each service that's running,
so running nova-conductor in this test suite added more info here.

Note that this uses the previously-added ping() call in conductor to
determine when the service is available. The compute manager pings the
conductor every ten seconds for ten attempts and then falls back to the
default configured RPC timeout. This should be a reasonable compromise
between requiring strict service startup ordering and extreme delays.

Part of blueprint no-db-compute.

Change-Id: Ie2953f7ae79819a1b6e24e8997ed4332fd4d2356

0 files changed, 0 insertions, 0 deletions