hyperconverged pve: To Upgrade or simply stay with a running system

rainer042

Well-Known Member
Dec 3, 2019
37
3
48
124
Hello,

I would like to ask you how you deal with updates for pve and hyperconverged ceph? I am now considering to update a productive pve 7.2 cluster with 12 hosts running Octopus. With an update to 7.3 I also would have to update the hosts cluster to at least ceph pacific. There is a detailed description how to perform the pve as well as the ceph update and I did already quite often updates of pve itself which is a process that can be done host by host. The ceph upgrade however is a kind of all in one. And if it should fail, the cluster could stop working and all VMs on it and the services they provide would also stop. A desaster scenario. There are file backups of quite everything, but to get everything up and running again would probably need days of work. So better do not update ceph?

The alternative of upgrading are:
1 stay with ceph octopus (no update) and only update to pve7.3.
2. do not update anything, just let the system run
3. upgrade pve and ceph

So I would simply like to know how you handle this dilemma between upgrading and availability for your systems? Or are your experiences such that you would say: no problem at all with choice 3.
 
Proxmox instructions on upgrading Ceph are designed and writted with hyperconverged cluster in mind. Every ceph upgrade I've done on Proxmox clusters allow both versions to run toghether meanwhile you are installing updated packages and restarting services. If needed, specific settings are detailed for each upgrade step. Issues may arise, so plan for some downtime when you run the upgrade.

If you've never done it before, I recomend that you install a test cluster either in hardware or as VMs and practice the upgrade procedure before doing so with your production cluster.
 
If you have the time and storage to do full backups, do a clean install. Less issues this way. Can backup the VMs to Proxmox Backup Server.

Promxox Ceph allows you to update each node a time. Just following those instructions carefully.