[PVE-User] Feedback on 4.x to 5.x upgrade documentation

Mark Schouten mark at tuxis.nl
Mon Jan 29 17:02:51 CET 2018


today I upgraded a 4.x cluster with Hammer (Ceph) to 5.1. I had almost no 
issues thanks to documentation and quality software.

I first upgraded Hammer to the latest hammer (older pveceph versions installed 
a static apt-repo to gitbuilder stuff) and restarted all mons, and then osds. 

Then I started the upgrade to Jewel, of which the documentation suggests you 
should do it per node. Which is different from what Ceph says; it suggests to 
do it per service (Mon, OSD). I think the reason to do it per host is that 
there is some chown'ing involved [1]. However, there is an easier (and imho 
less scary) way. [2] tells about an option to make the Ceph daemon test 
permissions and not drop privileges if it shouldn't. This allows for a default 
upgrade, after which you can stop and chown per OSD/mon, instead of per host.

The upgrade to Luminous was without any issues.

Then, upgrading to PVE 5.x, I ran into an issue which it seems is caused by 
KRBD. I was unable to migrate the hosts with KRBD enabled. I don't know why, 
and unfortunatly I cannot test it anymore (it was a production cluster so I 
had to keep going). But maybe this is something that can be debugged by 
someone in a lab or something. :)

Thanks for this good stuff!

[1]: https://pve.proxmox.com/wiki/Ceph_Hammer_to_Jewel#Set_permission
[2]: http://docs.ceph.com/docs/jewel/release-notes/?#upgrading-from-hammer
Kerio Operator in de Cloud? https://www.kerioindecloud.nl/
Mark Schouten  | Tuxis Internet Engineering
KvK: 61527076  | http://www.tuxis.nl/
T: 0318 200208 | info at tuxis.nl

More information about the pve-user mailing list