I have been reading about fencing:
https://www.ovirt.org/develop/developer-guide/engine/automatic-fencing/
https://www.slideshare.net/MartinPeina/host-fencing-in-ovirt-fixing-the-u...
Looking at Edit Cluster, I see:
X Enable Fencing
Skip fencing if host has live lease on storage
Skip fencing on cluster connectivity issues Threshold 50
i.e. Fencing is Enabled and neither of the two 'Skip' options are checked
Looking at Edit Host... Power Management, I see:
X Enable Power Management
X Kdump integration
Primary
[ ... and the remaining fields are populated with our ILO address & credential info
...]
OK, I get it now. Here's the story:
- ovirt-engine runs on Cluster A in Data Center A
- When ovirt-engine is unable to reach Cluster B in Data Center B, given enough
disruption, Fencing will kick in and try to mitigate the problem using various techniques,
including (eventually) power cycling via the ILO
One path forward for me is to check:
X Skip fencing on cluster connectivity issues Threshold
50
And twink with the Threshold to be more suitable for my cluster:
X Skip fencing on cluster connectivity issues Threshold
2
OK, I have a plausible model for understanding what has been happening.
Thank you for your assistance.
--sk