Updates:
Owner: [email protected]
Labels: -Type-Defect -Priority-High -Milestone-Release2.10
Type-Enhancement Priority-Medium Milestone-Release2.14
Comment #4 on issue 917 by [email protected]: Node evacuation fails if a
node in a different nodegroup is rebooting
https://code.google.com/p/ganeti/issues/detail?id=917
I've just had a quick look. It definitely is is a
feature request [helgav: just marking feature requests
as bugs won't help anybody, so don't do it!]; if a node
is not marked as offline, Ganeti is entitled to the
assumption that it can reach the node.
In a sense, Ganeti is even doing the only sensible thing.
The IAllocator protocol requires a description of the
cluster and a specification of the task.
We could try lying to the IAllocator and pretend the
cluster consisted only of the node group in question.
Besides being a major refactoring of the relevant code
(the code collecting the cluster data is shared among
all forming of IAllocator requests), it is also
conceptionally not as simple as it seems. While not a
desirable situation, it is allowed and handled by Ganeti
that instances spawn over node groups (primary on one
and secondary on the other). Such a situation can arise
during evacuation of a node group (as the way to move
an instance to another group is to first move the secondary
to the new group, migrate, move primary also the new
node group). As node evacuation also has to work while
a node group evacuation is in process, we have to deal
with this situation. The tricky aspect is that in that
situation, restricting to the nodes of the given group
doesn't cleanly result in leaving out instances from the
cluster description.
--
You received this message because this project is configured to send all
issue notifications to this address.
You may adjust your notification preferences at:
https://code.google.com/hosting/settings
