[pve-devel] Storage::Plugin::deactivate_volume not executed, when VM does shutdown

[Jasmin J.]

 > IMHO DRBD8 is old and replaced by DRBD9. Linbit will provide a DRBD9 driver
 > in future ....
Yes they will, but as stated in http://pve.proxmox.com/wiki/DRBD9
   "DRBD9 integration is introduced in Proxmox VE 4.x as technology preview."
and Linbit says (http://www.gossamer-threads.com/lists/drbd/users/28444#28444)
   "If you have a two node setup, stay with DRBD 8.4,
    you don't gain anything from the new features of DRBD 9,
    but, as you found out, still may be hit by the regressions."
So there are good reasons to stick with DRBD8 currently and this is what my
plugin is for. It will be very minimalistic in the first place and might be
used only a few months, until DRBD9 is stable.

 > First, we call deactivate if you migrate a VM to another node.
OK, I haven't seen this. I will test once my plugin is ready.

 > For HA, you need to fence the node before you start it on another node,
If I am using HA. Currently I have two machines and as a first step I plan to
do a kind of manual "HA". The automatic switch over requires a third machine
for quorum and I am not sure if I am able to get this running till end of next
week.

 > so there is no need to deactivate the volume at VM stop time.
It seems you insist the behaviour is good as it is, there is another
mechanism to cover it before migration and you don't want to change it.
Then it wouldn't make any sense to fix this and provide a patch, when it would
not be accepted. You are the project leader and you decide.

 > Besides, DRBD9 has built in support for auto-promotion, so it would
 > be preferable to use that mode (strongly suggested by Linbit to avoid
 > split brain).
You are so right, but it is not available in DRBD8 and DRBD9 is currently not
stable for production use. That's one of the reasons for me to provide a DRBD8
plugin which handles this good enough and to avoid the dual-primary mode. Yes,
then it might not be possible to do a live migrate (I need to test this). At
least it is not a requirement for me now and in case of HA it isn't either a
problem when one machine is dead (fenced) in a two node cluster.

 > So if the VM stops, all used DRBD9 resources changes back to
 > secondary - automatically!
As far as I remember my tests with DRBD9 two weeks ago this is true only, when
"deactivate_volume" is executed and the LVM volume is deactivate by
"lvchange -aln". But I didn't use the DRBD9 storage plugin with drbdmanage.
Instead I used a manually configured LVM VG on top of DRBD9.