Closed jabl closed 7 years ago
OK sounds reasonable :)
However, I think it would be better to keep managing the opensm service but set the service to disabled instead. I'll close this PR and add it manually.
Something like this:
rdma_opensm_enabled: "no"
The task:
- name: Manage the opensm service service: name: "{{ rdma_opensm_service }}" state: "{{ rdma_opensm_state }}" enabled: "{{ rdma_opensm_enabled }}"
Thanks, I guess that works too. :)
With lots of hosts, opensm can require a non-trivial amount of CPU time. So it shouldn't be run on a compute node where it can disturb batch jobs. Also, with lots of hosts, if all of them run opensm in standby mode I guess there will be a lot of polling on the network to check whether the master is still up.