Comment 10 for bug 1714248

Revision history for this message
John Garbutt (johngarbutt) wrote :

So idea for brute force fix...

On above change, we detect the conflict. We see if the name is duplicated, if it is we delete the old resource provider, to do that we probably have to delete all its allocations first, possible also its inventory, etc.

Once that is all done, we create a new resource provider, then add the appropriate allocations.

... But now we have a race where another instance will come in an try to get allocations on the new thing, just after we create it, but before we add back in the allocations. Boom, dang, the sky is falling down.

Now in the virt driver we know the node has the instance_uuid is set, so probably not a total disaster, but it seems a nasty race.

... not sure where to go with this.