How power off node?

[Penobscot]

Forgive me if this question is asked and answered often. I've tried reading the forum and manuals, and asking AI, and am more confused than ever.

I come from a VxRail background. That technology uses vSphere and vSAN. Shutting down a node requires migrating running VMs elsewhere (either manually or via DRS), putting it in maintenance mode, and powering it off from vCenter or the iDRAC. Basically, a few mouse clicks. Taking it out of maintenance mode is roughly the reverse.

Earlier this month I spun up a three-node cluster on Ceph after having used a standalone PVE for several months. All the normally running guests are configured for HA. Ceph has three replicas. I set the shutdown policy to migrate.

I've found ha-manager crm-command node-maintenance enable <node> will migrate the guests off the node (and disable will migrate them back on). Then what? Specifically, is there anything I need to do with Ceph before or after maintenance? I'm familiar with ceph -s, but otherwise I'm lost.

I need to do hardware maintenance on all three nodes and want to do one machine at a time. Thanks.

 

[robertlukan]

Hi,

Make sure you have healthy cluster first and you don't have some major load ongoing(like backups, patching, upgrades on VM-level), and I would do it in this order:

1. disarm HA in PVE (if you have it enabled)
2. migrate all VMs from host you would like to power off (yes, I like this process manually), maybe it works with ha-manager maintenance mode but I have not tried it yet
3. in Ceph enable no-out flag, Ceph will display warning, it is a global flag
4. stop Ceph osd-s and monitor, systemctl stop ceph-osd.target and systemctl stop ceph-mon.target on a node you want to poweroff
5. do whatever you need to do ... reboot/poweroff
6. after node comes back up, it will start osd-s and monitor, and ceph should heal by itself, make sure to check the status before you continue with a next node

After you are done with all three, remove no-out flag.

I hope this helps, otherwise ask