DRBD Sanity Check Please

It’s time for $work to upgrade from a single mail server to an active/passive mail server cluster. (Presently we’re using postfix and dovecot. I don’t see a need to change that part of the stack.)

The idea I have is to build a pair of servers with LVM logical volumes for the mail spools. On top of the logical volumes, run DRBD, and format that with XFS. Those would be some combination of enterprise SSDs.

The third element is to have a backup server with a bunch of big spinning rust drives, running a ZFS pool. One of the filesystems would be an identically sized zvol with DRBD on top of it. I’m assuming this will provide a way to snapshot the zvol for replication to other ZFS backup servers further offsite.

Is this going to work?

1 Like

I don’t see why it will nor work but the biggest issue will be skillset of the people that will maintain this and support this when you are not around. Just makes sure that you have buy in and leave a good passdown log of the build process and how they backup/snapshot system works.

1 Like

That’s a super relevant point. My effort involves combining build notes in Confluence, and a series of Ansible playbooks to help with recovery.

The actual building of the LVM volume is by far the most demanding task: searching for LVM commands can easily give you bunk. Luckily, I accept my new AI master’s search summaries…
…But I anticipate being around for another decade :slight_smile:

1 Like

I’m still trying to clear my plate to get to this project. So much daily drudge to clear out of the way :frowning: