Complex systems do not always fail for complex reasons. They quite often fail for the absolutely dumbest possible reasons.

- nelhage

Like say, for instance, when you’re debugging why one of your servers takes 7 seconds to do what the other server can do in less than 1.

root@black-mesa:~# time lvcreate -n test -L 1G xenvg
 Logical volume "test" created

real    0m0.309s
user    0m0.000s
sys     0m0.008s
root@torchwood-institute:~# time lvcreate -n test -L 1G xenvg
 Logical volume "test" created

real    0m7.282s
user    0m6.396s
sys     0m0.312s

Sometimes it turns out to just be that the directory it’s logging to takes 7 seconds to list:

root@black-mesa:~# time ls -a /etc/lvm/archive/ >/dev/null

real    0m0.005s
user    0m0.000s
sys     0m0.004s

root@torchwood-institute:~# time ls -a /etc/lvm/archive/ >/dev/null

real    0m7.007s
user    0m6.644s
sys     0m0.364s

And occasionally, that’s not just because your disk is failing or you’re running into caching issues. Occasionally, it’s just because that directory somehow has hundreds of thousands of files in it:

root@torchwood-institute:~# ls -a /etc/lvm/archive | wc -l
301369

And very, very rarely, if the gods are smiling on you, deleting those hundreds of thousands of files causes things to work again.

root@torchwood-institute:~# find /etc/lvm/archive -name '.lvm_torchwood-institute.mit.edu_*' -delete
root@torchwood-institute:~# time ls -a /etc/lvm/archive >/dev/null

real	0m0.015s
user	0m0.000s
sys	0m0.012s
root@torchwood-institute:~# time lvcreate -n test -L 1G xenvg
  Logical volume "test" created

real	0m0.341s
user	0m0.000s
sys	0m0.016s
root@torchwood-institute:~# time lvremove -f /dev/xenvg/test
  Logical volume "test" successfully removed

real	0m0.226s
user	0m0.004s
sys	0m0.012s

It kind of sucks to spend a week on and off, talking with developers, trying to figure out what’s going on. But it’s also really nice when it turns out to be something I can just fix myself.