Red Hat Bugzilla – Bug 1462111
HA LVM using CLVM fails to relocate due to failed exclusive activation on the passive node
Last modified: 2017-09-27 16:38:02 EDT
This really goes deeper than the agent, and I'm not sure that delays as proposed would be a good path forward here (as it would likely create some form of impact to use cases not affected by this).
Ultimately this is a problem with clvmd, in that it can't handle a request when there is a member which hasn't started yet. Having the agent need to accommodate this behavior doesn't seem like the best target.
It is an unfortunate timing window for sure, but do we have the ability to work around this with timings or delays elsewhere? Something that delays the node joining for the amount of time needed to get everything recovered elsewhere?
As we're looking at 6.10 planning, it would be great if we could find a way to eliminate this from consideration, as it would have minimal impact to our overall customer base and could be too invasive for prod-phase 3.
Let us know if we can pursue these other angles.