Please can you be more specific about what feature do you want/need exactly? I am confused.
Proxmox 4 HA VM Freeze State
- Thread starter adamb
- Start date
Please can you be more specific about what feature do you want/need exactly? I am confused.
A simple toggle to determine how HA reacts on a node reboot.
I did know about shutdown, but we don't want to shutdown our front ends, we just want to reboot and have the VM's fail over. Not to mention, most of our clusters and clients are remote, so its not like I can just walk in the server room and power up the server after its been shutdown.
Is there already a feature request on bugzilla.proxmox.com? If not, please can you create one?
Thanks Dietmar, I missed that point sorry.
This means that if a VM is under HA control and running on a node I'm about to shutdown (using the GUI button or as well init/poweroff/shutdown commands ?), it will be migrated to an eligible node, in the HA group, by default ?
yes
Yes.After node shutdown, VMs will get started on other nodes.[/QUOTE]
Started after shutdown != Migrated before shutdown
I'd gladly prefer to keep the resource(s) online if possible.
We're a few weeks into to trialling Proxmox and I just found this thread because HA isn't doing what we think it should.
Belokan, like you and other posters, we want our VMs to stay online.
Started after shutdown != Migrated before shutdown
I don't care if a physical server is rebooted or shutdown, I want the VMs to stay online if possible.
XenServer gets this right, if you reboot, shutdown, or put into "maintenance mode" a host from XenCenter it pops up a dialog asking you want to do with each VM it's hosting.
Whilst also having optional HA to start VMs if hosts crash or whatever.
I believe HA will maintain a limit to resource usage so that at least one node can be lost and still run all VMs as well.
Oracles VM platform (OVM) does a similar thing.
Enable maintenance mode on a host at the VMs migrate away based both on which servers are least-used and anti-affinity groups (groups of VMs that should not share a single physical machine).
IIRC both OVM and XenServer won't be graceful if you reboot or shutdown from the command line, but don't hold me to that.
I would be fine for Proxmox to only migrate gracefully if reboot or shutdown were selected from the web GUI or API, and be brutal to CLI users.
Mike
HA is high availability and not fault tolerance, what you mean is fault tolerance - so PVE is doing it well, you think wrong here.
https://en.m.wikipedia.org/wiki/High-availability_cluster
https://en.m.wikipedia.org/wiki/Fault_tolerance
This is possible but it might be a very difficulty implementation.
But of course and feature like an maintenance mode would be nice. So you do not have to evacuate the node, proxmox will do it and after this the VM from before should fail back on the origin node.
Good news is, the dev's have made a change. Its still in testing but we will have it soon.
https://bugzilla.proxmox.com/show_bug.cgi?id=1378
https://bugzilla.proxmox.com/show_bug.cgi?id=1378
I've checked the patch documentation and it's still based on "recovering" the service, not failing over the service is case of reboot/shutdown.
You can work around this for now with a ha-group, add all desired services and all nodes.
If you want to shutdown/reboot a node just remove it (or lower its priority) from the group and all HA services will move to another node automatically, not ideal yet but can help automate this, especially if scripted.
Hi Thomas,
Thanks for your answer, that's how I manage maintenance right now indeed. But sooner or later, the failover before reboot/shutdown request will come back as it's available on most clustering/HV products.
I can't believe a "dummy" philosophic point like HA != FT avoid such looks-like-to-be-easy-to-implement feature to be added to Proxmox
Cheers,
Olivier
Thanks for your answer, that's how I manage maintenance right now indeed. But sooner or later, the failover before reboot/shutdown request will come back as it's available on most clustering/HV products.
I can't believe a "dummy" philosophic point like HA != FT avoid such looks-like-to-be-easy-to-implement feature to be added to Proxmox
Cheers,
Olivier
I just tested this after the latest upgrade and it works as advertized! Here is a cookie, and we will give Thomas a cake!
Thanks all involved!!
(PS. I have complained about this very same situation since day 1 on proxmox but was never put my finger on what the fix needed to be, so am a happy camper now! DS.)