I’ve been recently testing Urbackup with btrfs, since my ext4 partition of 8TB and a gazillion files seemed to cause some issues for Urbackup. Results are dreadful…
It performed well, for a week or two. And then, after some snapshots (I have 35 mailservers backing up), it’s terrible. I’ve tried different mount options, but nothing seems to help.
Now, there is ZFS, which works far better and is more stable than btrfs. It only lacks reflinks. But why is a reflink so important? I could care less if reflinks are enabled or not. Storage is not that expensive, performance is.
So please, implement ZFS snapshots, disable reflink if you detect ZFS. Both btrfs and ext4 seems unwilling for larger installs…
Snapshots and reflinks work differently. You can’t dereference a single file in a snapshot, you have to discard the entire snapshot, snapshots are (except in BTRFS) immutable.
Long term it’s also pretty wasteful to keep n snapshots vs versioning individual files. What you’re describing is no different than rsyncing your systems and creating regular snapshots of your backup destination. It’s doable but different.
I do not see how x zfs snapshots are more wasteful than keeping x versions of all files at a given moment.
I was rsyncing to zfs, which works like a charm, very efficient and fast. But I like the way urbackup autoconfigures, has restore-features that are better, and better management overal.
But I’m now struggling with millions of files and filesystems that can handle the rather inefficient ways that urbackup handles ‘snapshot-like’ functionality. I get where it comes from, ten years ago, but currently we have ZFS to fix the inefficiency.
urbackup also works on windows server , ie without zfs. and typically works with most fs.
If there were more devs what could be implemented is multiple storage backed, and make a specialized zfs one , an lvm one, a btrfs one, taking advantages of the functionalities.
Unfortunately right now there s a single dev