Best distributed filesystem for CentOS 7?

[Leek]

I've been pulling my hair trying to set up a redundant e-mail server out of one CentOS 7 node and two modern Debian-based nodes (one Debian and one MX Linux). I can probably figure out the e-mail part if and whenever I get to it, but the distributed storage part is kicking my butt. Nothing seems to work for me across these two distributions.

First I tried GlusterFS: The Debian nodes have Gluster versions 9 and 10, but the newest I could find for CentOS 7 was Gluster 7.8, which is incompatible with the newer versions.

Then I tried Ceph: This was the most confusing. All of the tutorials I found say to do completely different things. I made some progress with ChatGPT but she eventually got stuck walking me around in circles. I just have no clue how to use this program and I couldn't find any noob-friendly guide anywhere. When I asked their community for help I was told DRBD or ZFS may be better options for my use case, which is what led me to...

...giving DRBD a shot: I thought this would work after I got all of the nodes on the same version, but then I realized that version (drbd 8) only supports two nodes while my setup requires three, a feature which apparently was introduced in drbd 9. CentOS isn't the problem here - it is actually the Debians I couldn't find a damned drbd 9 kernel module package for. Building a kernel module from scratch is not something I'd trust myself to do successfully, specially after reading the build guide for it.

I know I'm not the only CentOS user that's ever needed something like this so I beg for your assistance. Buying a fourth cheap node is an option for my wallet but not for my pride. I know CentOS is dead blah blah blah - tell that to my torrent client that's still seeding it regularly. Dead or not, it should be able to do something like this, right?

[TrevorH]

How about not using CentOS 7? It only has until June next year and then it dies anyway so you will be repeating the build in less than a year in order to stay secure. There are rebuilds of both RHEL 8 and 9 by Alma, Rocky and Oracle, all of which try to be as similar to the RHEL upstream as possible.

I doubt if DRBD is the solution you are looking for as you can only use the DRBD device on one system at a time unless you also use a cluster based filesystem like GFS2. The advice generally given about doing that is that if you need it then you probably have a design problem that you should solve instead! It's slow, not exactly reliable and when it goes wrong it goes really wrong.

If I needed something like this I would probably do 2 node DRBD with HA failover and use that to create an NFS share that can be used by as many other systems as you want.

[Leek]

How about not using CentOS 7? It only has until June next year and then it dies anyway so you will be repeating the build in less than a year in order to stay secure. There are rebuilds of both RHEL 8 and 9 by Alma, Rocky and Oracle, all of which try to be as similar to the RHEL upstream as possible.

I doubt if DRBD is the solution you are looking for as you can only use the DRBD device on one system at a time unless you also use a cluster based filesystem like GFS2. The advice generally given about doing that is that if you need it then you probably have a design problem that you should solve instead! It's slow, not exactly reliable and when it goes wrong it goes really wrong.

If I needed something like this I would probably do 2 node DRBD with HA failover and use that to create an NFS share that can be used by as many other systems as you want.

So DRBD isn't good at all...? Damn it. This really, really sucks. I thought this would be a piece of cake for Linux in 2023 but I was dead wrong. I thought I should be happy because I finally got DRBD working (at least on the two Debians) but apparently not.

Do you think I should avoid DRBD entirely and just use Gluster instead? Will that work? I need all servers to be the same, not just two holding the master data. I may be able to get Gluster working if I just stick to Debian, but at this point I'm not even sure that's worth it. I'm not sure anything is worth it anymore. It's almost like nobody on the internet ever tried to set up a friggin cluster because all the information I find is wrong. At this rate I'll die with a gmail or hotmail account left on my grave as contact information...

[TrevorH]

I do not understand why you want to use a nearly EOL distro like CentOS 7 for this. It's coming up for retirement and due to draw its pension and is looking forward to a nice life doing nothing. If you want to use a RHEL-alike o/s, use one of the newer rebuilds. CentOS is dead.

[Leek]

I do not understand why you want to use a nearly EOL distro like CentOS 7 for this. It's coming up for retirement and due to draw its pension and is looking forward to a nice life doing nothing. If you want to use a RHEL-alike o/s, use one of the newer rebuilds. CentOS is dead.

I just wanted to take advantage of an existing system that happened to be running CentOS, but you're right, this use case does require better security than its current job and a non-maintained distro won't cut it. I'd still rather get a third Debian system than having to reconfigure this system from scratch. I'll consider it an admission fee into the CentOS forever loyalty club. Somebody's gotta go down with the ship and this happens to be a perfect candidate.

(Well, forever or until the newest LTS Java stops working on it, whichever comes first )