Unhardcode live_migration_tunnelled
As we do not have shared instance storage configurable for computes,
live migration with block migration is the only live migration
type choice for VMs not booted from cinder volumes.
At the same time we have uncoditionally enabled
"live_migration_tunnelled" option in nova.conf that is NOT
compatible with block migration (see help string for that option).
This can be conditionally enabled in the future when MCP
supports e.g. Ceph as a backend for shared instance storage
on computes.
This patch picks default value from nova service unless it is
configured via compute:libvirt:live_migration_tunnelled option.
Starting from Newton is disabled by default.
https://github.com/openstack/nova/commit/61f122637b8c9952e28983de81638941dc4e7bc4
Change-Id: If2d2e6f782f79dc45655091a7615d884ddf3d999
Related-Issue: PROD-15929
diff --git a/nova/files/ocata/nova-compute.conf.Debian b/nova/files/ocata/nova-compute.conf.Debian
index 54708a3..337b3f5 100644
--- a/nova/files/ocata/nova-compute.conf.Debian
+++ b/nova/files/ocata/nova-compute.conf.Debian
@@ -6228,7 +6228,9 @@
# 'block_migration_flag' to enable tunneled migration.
# (boolean value)
#live_migration_tunnelled=false
-live_migration_tunnelled=true
+{%- if compute.libvirt.live_migration_tunnelled is defined %}
+live_migration_tunnelled = {{ compute.libvirt.live_migration_tunnelled }}
+{%- endif %}
#
# Maximum bandwidth(in MiB/s) to be used during migration.
diff --git a/tests/pillar/compute_cluster.sls b/tests/pillar/compute_cluster.sls
index e200640..824a569 100644
--- a/tests/pillar/compute_cluster.sls
+++ b/tests/pillar/compute_cluster.sls
@@ -67,6 +67,7 @@
port: 11211
libvirt:
hw_disk_discard: unmap
+ live_migration_tunnelled: False
upgrade_levels:
compute: liberty
libvirt_service_group: libvirtd