Proxmox Cluster from 5.1 to 5.4, eventually to 6.0

[Richard Leishman]

Hello.

I am looking to upgrade a production cluster of 8 nodes without restarting the virtual machines inside them, I do not use any form of Proxmox failover/fencing management and storage is NFS completely separate to the cluster. I read on the Proxmox docs that to get to 6.0, I first need to get from 5.1-41 to 5.4 and run the test tool.

Does anyone know that if I was to dist-upgrade one machine at a time and migrate the VM's, will this cause any issues to the cluster?

Thank you in advance.

 

[aaron]

If you migrate all VMs off the node that you are about to upgrade it should work fine.

Plan it so that you will always migrate to a node with the same or a newer version of PVE.

Regarding upgrading to PVE 6: check the wiki page[0]. If you follow those steps it should work fine.

[0]: https://pve.proxmox.com/wiki/Upgrade_from_5.x_to_6.0

 

[Richard Leishman]

Thank you, do you think I may come across any issues when adding a node running pve 5.4 to a 5.1 cluster?

 

[RokaKen]

Thank you, do you think I may come across any issues when adding a node running pve 5.4 to a 5.1 cluster?

I don't recall any specific issues with 5.1 -> 5.4, but browse the Roadmap release notes WRT your configuration. Also, check your corosync.conf and ensure you are using IP addresses for ring addresses. That should be the case, but prior to 5.1 the cluster could have been created using host names like:

nodelist {
  node {
    name: pve-01
    nodeid: 1
    quorum_votes: 1
    ring0_addr: pve-01
  }
  node {
    name: pve-02
    nodeid: 2
    quorum_votes: 1
    ring0_addr: pve-02
  }
...

That will continue to work in 5.x and upgrades will not change it automatically, but it depends on getaddrinfo returning the correct address every time. That means configured /etc/hosts and DNS servers MUST be consistent as getaddrinfo can return multiple addresses.

Now would be a good time to change that to specific IP addresses like:

nodelist {
  node {
    name: pve-01
    nodeid: 1
    quorum_votes: 1
    ring0_addr: 10.1.1.101
  }
  node {
    name: pve-02
    nodeid: 2
    quorum_votes: 1
    ring0_addr: 10.1.1.102
  }
...

See the relevant sections of Cluster Manager documentation and how to properly edit the corosync.conf, if required.

 

[Richard Leishman]

Thank you both.

 

[Richard Leishman]

Hello,

Just thought I would give a little update to this as I hate leaving unsolved threads.

I took the advice above, one host at a time. I moved all VM's to a spare host, used the "apt update && apt dist-upgrade" commands to update the host from 5.1 to 5.4 which went smoothly. The cluster ran perfectly fine with a mix of 5.1 and 5.4 nodes during the process.

I had an issue with one or two hosts where /etc/apt/sources.list was incomplete and only updated to the latest version of 5.1 but after getting the file right, this worked perfectly.

I will leave 6.0 for now, but when the time comes I think I will do a fresh install for that.

Thank you for your help.