HA for ceilometer agents/notifier not implemented
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack-Ansible |
Fix Released
|
Low
|
Miguel Alejandro Cantu | ||
Liberty |
Fix Released
|
Low
|
Miguel Alejandro Cantu | ||
Trunk |
Fix Released
|
Low
|
Miguel Alejandro Cantu |
Bug Description
When running multiple central-agents without workload partitioning, each agent will poll the same service and will produce duplicate samples[1].
Currently OSA does not implement the HA solution for the central agent nor the notifier(notifier can do workload partitioning as well). OSA also documents the ability to run HA central agents at the moment[2], which is unsupported with the current implementation.
The fix would be to add the bits necessary to deploy ceilometer central agents/notifier using Tooz with a supported backend solution[1].
[1] http://
[2] https:/
Changed in openstack-ansible: | |
assignee: | nobody → Miguel Alejandro Cantu (miguel-cantu) |
Changed in openstack-ansible: | |
status: | New → In Progress |
One way would be to write OSA to deploy a zookeeper ensemble( muli-server zookeeper cluster). It could use external ansible roles to do so, and it would only need to do so when len(groups[ 'ceilometer_ agent_central' ] is > 1. role-requiremen ts.yml?
Some thoughts arise, however:
* Do we run the zookeeper services in their own containers?
* If so, we should define a new environment in env.yml. That way the zookeeper services can run on their on infrastructure if need be.
* If we do use external ansible roles, is it appropriate to just add them to ansible-
Maybe we can talk about this during the weekly IRC meetings. I hate having to make these decisions by myself. Zookeeper seems like a pretty hefty application to deploy along side ceilometer, so I want to make sure I'm doing it right.