I know I like to dunk on #btrfs, but I literally had another outage due to a full disk. Today, a new behavior appeared: the mount switched to read-only and would not remount as writable without a reboot.

I was literally SSH'ing in to move half the data to a new ZFS pool when I saw I couldn't because everything was read-only. Thankfully, there was a recovery path:

doas btrfs subvolume list .|grep "snapshot\$"| awk '{ print $9 }'| doas xargs btrfs subvolume delete

0 days from the last btrfs-induced outage.

This time: "No space left on device" errors; df shows 518GB available. df -i shows unlimited inodes. Why out of disk space then? I ran out of "metadata" space. Fixable with: `btrfs balance start -dusage=0 /`

It's like defrag from DOS 6.2.

I look forward to wiping #btrfs out of my infrastructure. I haven't had a filesystem this needy since FAT16.

I tried tuning various parameters but after some reading came to the conclusion that lots of small files with very little RAM is about the worst case scenario for XFS.

#linux #snac #btrfs