Richard Gass wrote:
Hi Michael. Yes you are correct that the lv disk should be in use even
during reboots. I understand that you want to blast everything that is
not in use to cleanup unused lv's due to improper exits of the
vm/nodemanager. For me it makes me a little uncomfortable allowing
the fate of all the lv's to be decided at every change in vm state.
Also, what happens during migration.
Sure, and I don't disagree with the change. I just wanted to note why I
did it that way, and let you know that there is a risk of unused space
remaining claimed.
I can see arguments either way. If we blast everything every time, we
are pretty sure that the disk shouldn't have unused disk space for
extended periods granted high vm turnover on the host. However, it
makes me nervous doing this and also may have issues during migration,
suspends, etc.
Shouldn't be a problem during migration and suspends, because the LVs
are only released once the process exits. I'd be more afraid of someone
using the scratch VG to put other stuff on and find it blown away.
The other way we may need to have a separate thread running to check
the disk for unused vms and keep track. It maybe a nice place to put
all lv related stuff.
I think this is probably the best, but somewhat complex to implement.
I'll work on removing rpyc first before I work on a scratch vacuumer.
Greetings,
Michael.
