3.4 -> 4.0 with Ceph Server installed

bkraegelin

Renowned Member
I'm running a three node 3.4 cluster with integrated ceph server.

Upgrading another test cluster showed that one has to be very careful as you have to create a new cluster. So I have some questions:

  • Did anybody already upgrade a running 3.4 ceph cluster to 4.0?
  • Will I loose my ceph configuration?
  • What is the right order of steps to not fully loose the ceph cluster and all its OSDs (if possible)?
Thanks for any help.
Birger
 
I'm trying to find this out also. I don't really have a way to test. Anyone completed the upgrade from 3.4 to 4.0 with Ceph that can provide some insight?
 
I'm running a three node 3.4 cluster with integrated ceph server.

Upgrading another test cluster showed that one has to be very careful as you have to create a new cluster. So I have some questions:

  • Did anybody already upgrade a running 3.4 ceph cluster to 4.0?
  • Will I loose my ceph configuration?
  • What is the right order of steps to not fully loose the ceph cluster and all its OSDs (if possible)?
Thanks for any help.
Birger

Hi Birger!

P.S.
I previously did a full backup for all VMs located in the CEPH-storage, but I did not come in handy.
:)

---
Best regards!
Gosha
 
Last edited:
- upgrade to hammer first on proxmox 3.0/wheezy. (use ceph.com hammer wheezy repositories)
-restart all vms or migrated them, to run them on last librbd
-restart mon
-restart osd.

Then
upgrade proxmox nodes to 4.X (change also ceph.com hammer wheezy repository to hammer jessie repository)
 
Definitely doable - I somewhat broke a Proxmox node acting as Ceph mon and another Prox node that is host to OSD's. Took a bit of trial and error, but we successfully upgraded and had no loss of config/data. Note: this wasn't a production cluster when this happened, thankfully!
 
i did a reinstall of my ceph nodes. Make sure to back up /var/lib/ceph for mon,osd and mds databases and keys. Osd's keep their keys on the disks and are mounted via udev. You could even move your osd disks to another node and they should work fine.
I had no downtime and ceph did not lock up