Upgrading from CentOS 7.6 to AlmaLinux 9.3 Without Format Disks

[Gr33n R10t]

Dears,

I have to upgrade my BeeGFS cluster from the version CentOS 7.6 to

AlmaLinux 9.3.

There is a way to unmount the disks and do the upgrade to a new distro

without losing any data?

Or I have to re-create all the cluster from scratch copying again all the data?

Thanks in advance for any suggestion!!

Bests

g.

[Michael Robbert]

Hi all,

I'm in a similar situation. Getting ready to upgrade our cluster from CentOS 7.9 and we're planning to move to Rocky 9.3. Since we've been running CentOS 7 we've been stuck on the BeeGFS 7.2.x train and plan to move to 7.4.latest. Looking through the documentation I don't see any clear OS upgrade tips or recipes. Based on the upgrade guide (https://doc.beegfs.io/7.4.3/advanced_topics/upgrade.html) it looks like it should just work if you add "Upgrade OS" between steps 3 and 4. I have too much data to make it practical to backup the whole filesystem (it is a scratch filesystem so users no that there are no backups), so I plan to just do the "Minimal Backup of Important System Configuration" found in the backup link on that page. It does seem a little scary to do a fresh install on all 6 of my BeeGFS servers without an easy way to test the process or have a rollback plan in case it doesn't work. I may look at doing just an OS back of each host or since my OS disks are all mirrored I may try splitting the mirror and installing the new OS on just one disk, then reassemble the mirror once I know it is working. 

I would be very appreciative if anybody from the community could share how they have handled major OS updates on BeeGFS servers in the past. Any tips, tricks, or gotchas to look out for?

Thanks,

Mike 

[Andreas Skau]

Hi,

What I did was basically what you're saying. IIRC the procedure ended up being roughly: stop all the clients, give the storage and meta servers a good spring cleaning (beegfs-fsck, dispose unused, xfs.fsck, zfs upgrades, etc etc), then when everything is as good as possible: backup, reinstall, re-tune, remount the volumes and start to spin up the daemons. 

Quoting myself from last year:

Thu, Apr 27, 2023, 12:44 PM
to fhgfs-user

Hi Olafur,

I've just done a CentOS 7 to Rocky 9 migration. I believe we were on BeeGFS 7.2.6 and went to 7.3.3. We have 4 metadata servers with zfs, 12 storage servers with xfs. This is on ConnectX5 FDR ib. No issues of note, except that the recommended nr_requests tuning of 4096 is no longer supported, so you have to reduce that to 2048.

Can't see any reason to go via an EL8-based installation.

Just back up your /etc folders, reinstall the servers (make sure you don't format the wrong drives by accident), install beegfs as before and apply the OS tuning as before. Also a good idea to re-read the storage and metadata tuning guides just to check that nothing was missed...

Good luck!

Andreas

--
You received this message because you are subscribed to the Google Groups "beegfs-user" group.
To unsubscribe from this group and stop receiving emails from it, send an email to fhgfs-user+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/fhgfs-user/d6e942aa-3f3d-4cf9-af75-cb11d734bb64n%40googlegroups.com.