Ensure <service>_restart_bundle do not run concurrently

Now that paunch containers are run asynchronously, some
containers used for HA orchestration can run concurrently,
which triggers a deadlock and cause them to time out and
exit in error.
To overcome this concurrency deadlock, tweak the start_order
configuration of each HA service to make sure that no
<service>_restart_bundle container can be started at the same
time during a given step.

Change-Id: I9ae55101945978ea6c65d37eaa8221ed3d96a7f5
Closes-Bug: #1873893
This commit is contained in:
Damien Ciabrini 2020-04-23 16:42:46 +02:00
parent 07e9f470d8
commit 6901246827
5 changed files with 5 additions and 5 deletions

View File

@ -198,7 +198,7 @@ outputs:
# update (scale-out, etc.)
TRIPLEO_DEPLOY_IDENTIFIER: {get_param: DeployIdentifier}
cinder_volume_restart_bundle:
start_order: 1
start_order: 2
config_volume: cinder
detach: false
net: host

View File

@ -180,7 +180,7 @@ outputs:
# update (scale-out, etc.)
TRIPLEO_DEPLOY_IDENTIFIER: {get_param: DeployIdentifier}
manila_share_restart_bundle:
start_order: 1
start_order: 3
config_volume: manila
detach: false
net: host

View File

@ -225,7 +225,7 @@ outputs:
# update (scale-out, etc.)
TRIPLEO_DEPLOY_IDENTIFIER: {get_param: DeployIdentifier}
rabbitmq_restart_bundle:
start_order: 1
start_order: 5
config_volume: rabbitmq
detach: false
net: host

View File

@ -225,7 +225,7 @@ outputs:
# update (scale-out, etc.)
TRIPLEO_DEPLOY_IDENTIFIER: {get_param: DeployIdentifier}
rabbitmq_restart_bundle:
start_order: 1
start_order: 6
config_volume: rabbitmq
detach: false
net: host

View File

@ -231,7 +231,7 @@ outputs:
- {get_param: RabbitCookie}
- {get_param: [DefaultPasswords, rabbit_cookie]}
rabbitmq_restart_bundle:
start_order: 1
start_order: 4
config_volume: rabbitmq
detach: false
net: host