about summary refs log tree commit diff
path: root/users/wpcarro/website/blog/posts/nixos-disk-full-note.md

Background

Every now and then NixOS hosts runs out of disk space. This happened to my IRC server recently...

No problem. Let's free-up some space with Nix's garbage-collection:

  • me
λ nix-collect-garbage -d # failed due lack of disk space

Ironically Nix needs to do an SQLite transaction before deleting stuff and SQLite can't do that if there's no space. This is especially funny because the SQLite is probably a DELETE.

Solution

First let's verify that our disk is indeed at capacity:

λ df -h
Filesystem                Size  Used Avail Use% Mounted on
devtmpfs                  399M     0  399M   0% /dev
tmpfs                     3.9G     0  3.9G   0% /dev/shm
tmpfs                     2.0G  3.7M  2.0G   1% /run
tmpfs                     3.9G  408K  3.9G   1% /run/wrappers
/dev/disk/by-label/nixos  9.9G  9.9G    0G 100% /
tmpfs                     4.0M     0  4.0M   0% /sys/fs/cgroup
tmpfs                     797M     0  797M   0% /run/user/0

Looks like /dev/disk/by-label/nixos is at 100%. Now let's find some easy targets to free-up space so that we can run nix-collect-garbage -d...

λ du -hs /* 2>/dev/null
8.0K    /bin
12M     /boot
0       /dev
200K    /etc
68K     /home
16K     /lost+found
9.0G    /nix
0       /proc
1.2M    /root
2.9M    /run
4.0K    /srv
0       /sys
44K     /tmp
12K     /usr
1.2G    /var

Okay: /var looks like an easy candidate. Let's recurse into that directory:

λ du -hs /var/*
40K     /var/cache
12K     /var/db
4.0K    /var/empty
4.0K    /var/google-users.d
211M    /var/lib
0       /var/lock
918M    /var/log
0       /var/run
4.0K    /var/spool
44K     /var/tmp
λ du -hs /var/log/* # /var/log looks promising
60M     /var/log/btmp
82M     /var/log/btmp.1
776M    /var/log/journal # ah-ha! journald. Let's clean-up some logs
8.0K    /var/log/lastlog
1.1M    /var/log/nginx
4.0K    /var/log/private
12K     /var/log/wtmp

To retain at most 1w's worth of logs:

λ journalctl --vacuum-time=1w

...or if you'd prefer to retain only 100M's worth of logs:

λ journalctl --vacuum-size=100M

Now Nix should be able to garbage-collect!

λ nix-collect-garbage -d

And lastly verify that it WAI'd:

λ df -h
Filesystem                Size  Used Avail Use% Mounted on
devtmpfs                  399M     0  399M   0% /dev
tmpfs                     3.9G     0  3.9G   0% /dev/shm
tmpfs                     2.0G  3.7M  2.0G   1% /run
tmpfs                     3.9G  408K  3.9G   1% /run/wrappers
/dev/disk/by-label/nixos  9.9G  5.1G  4.3G  55% /
tmpfs                     4.0M     0  4.0M   0% /sys/fs/cgroup
tmpfs                     797M     0  797M   0% /run/user/0

Closing Thoughts

Why doesn't Nix just reserve enough space to be able to GC itself? Not sure...