Shutdown applied to all nodes?!
- Thread starter magnayn
- Start date
So yes, it seems to be related to this.
But: we're not exactly using HA. We have a single VM (that is on proxmox1) set up as an HA resource as an experiment.
Everything else is VMs / CTs living on one or the other box using no HA features at all.
I can understand, perhaps, why that single resource might get offlined (but not really, given it was on the node that wasn't shut down), but it's very surprising to me that doing a "shutdown" on proxmox2 nukes every VM without warning, including ones that are just single VMs.
Why does it do that?
But: we're not exactly using HA. We have a single VM (that is on proxmox1) set up as an HA resource as an experiment.
Everything else is VMs / CTs living on one or the other box using no HA features at all.
I can understand, perhaps, why that single resource might get offlined (but not really, given it was on the node that wasn't shut down), but it's very surprising to me that doing a "shutdown" on proxmox2 nukes every VM without warning, including ones that are just single VMs.
Why does it do that?
when a node has a ha resource configured and loses quorum, it must fence itself to prevent the resource running twice.
also please check the docs regarding ha https://pve.proxmox.com/pve-docs/pve-admin-guide.html#chapter_ha_manager
also please check the docs regarding ha https://pve.proxmox.com/pve-docs/pve-admin-guide.html#chapter_ha_manager
Thanks for the docs link. We'll be combing over that as the plan is to upgrade to a proper 3-node cluster as soon as the box arrives.
I guess the behaviours that are surprising are:
- that it took down VMs and CTs that are not configured to be HA at all.
- that there is no warning in the UI. The person expected that "Shutdown" clicked on a node would take down just that node. I follow how it is the case that HA resources would also be taken down due to quorum rules. It was very unexpected that it also took out non-HA resources too.
I guess the behaviours that are surprising are:
- that it took down VMs and CTs that are not configured to be HA at all.
- that there is no warning in the UI. The person expected that "Shutdown" clicked on a node would take down just that node. I follow how it is the case that HA resources would also be taken down due to quorum rules. It was very unexpected that it also took out non-HA resources too.
Good to know; I presume though after a 'hard reset' those non-HA VMs still seemed to be down.
Would it not be possible to implement this so that doing this and taking down everything was not required?
At the very least, a warning in the shutdown dialog might be in order. I seem not to be the first to have been bitten by this..
Would it not be possible to implement this so that doing this and taking down everything was not required?
At the very least, a warning in the shutdown dialog might be in order. I seem not to be the first to have been bitten by this..
Ohh, just wait till you use pveceph purge and realize it wipes the entire CEPH production cluster. You then check the documentation once again to see if it was anywhere noted that it wipes the entire 30TB storage cluster. Nope... nowhere mentioned!
Subsequent pondering ensured.
pvecm and qm, and other commands which are proxmox proprietary commands are now banned from use in our system. Simply because we fear the documentation forgets to mention that using them will Nuke ever PVE cluster in the city with a 50km radius!
Subsequent pondering ensured.
pvecm and qm, and other commands which are proxmox proprietary commands are now banned from use in our system. Simply because we fear the documentation forgets to mention that using them will Nuke ever PVE cluster in the city with a 50km radius!
not that is has to do with the thread here but a quote from the manpage of 'man pveceph':
just fyi, the source code is fully open and nothing is "proprietary" : https://git.proxmox.com/