Hi Götz,
We’ve done a similar process which involves going from starting at CentOS 7 Nautilus and
upgrading to Rocky 8/Ubuntu 20.04 Octopus+.
What we do is start on CentOS 7 Nautilus we upgrade to Octopus on CentOS 7 (we’ve built
python packages and have them on our repo to satisfy some ceph-mgr things and such with
octopus on centos 7)
From here we have a process to migrate the node from CentOS 7 to Rocky 8 preserving the OS
and stuff, but you could also just reinstall the OS and reinstall ceph packages/config
files if need be.
Once on Rocky 8 and Octopus we then upgrade the ceph versions further.
The order of the upgrades would be like this:
CentOS 7 / Nautilus
CentOS 7 / Octopus
Rocky 8 / Octopus
Ubuntu 20 / Octopus
We aren’t changing from CentOS right to Ubuntu OS but process is similar enough.
The one time we did switch to Ubuntu we just did the same process then once on the latest
ceph version just reinstalled a node at a time from Rocky to Ubuntu. Probably are fine to
go right from CentOS 7 to Ubuntu but we figured it’d be more reliable to go from el8 rpms
to debs than el7 rpms to debs.
I would say make the priority matching ceph versions if possible, and then OS. This is
what has worked for us. Other people may have different experiences however.
In your case, your b choice is the closest to what we would do so I would say that should
be the safest.
Overall your a and b is kind of a mix of what we do for our upgrades.
Regards,
Bailey