How does one find which files are taking up 80% of the space on a Linux webserver?

Question

This is an actual cPanel server but the info here will probably relate to others so I thought it would make a good question.

The drive is constantly filling up. You've hunted down all the spare and random files you can. grep'd for coredump files and even removed some of the un-needed backups...

What would be your next move.

The actual server in question has 10GB of website files and the OS shouldn't take any more than 10GB so how do you track down what's filling a 50GB (virtual) drive?

Answer 1

I find ncdu (http://dev.yorhel.nl/ncdu) to be quite helpful for this.

Answer 2

Surely there are more elaborate ways, but the one I remember is

du --max-depth=1 -h /

Now take the directory that uses up most space (du --max-depth=1 -h /yourdir) and go deeper until you find your culprit.
If want your output sorted by size and don't care for the human-readable format, you could also do du --max-depth=1 /your_dir | sort -n

Answer 3

df -k shows which fs are the problem. Then cd to the top level dir for it and run du -xk | sort -n | tail -25 this will show the top 25 dir, sorted, for sun 9 or earlier, replace the x with a d.

Answer 4

Give gt5 a try.

Answer 5

I use the Gnome program baobab. You can run this on your desktop and t can connect via SSH to the server. It shows an easy to read graphical map of disk space usage. It's installed under Gnome as "Disk Usage Analyzer"

Answer 6

Note that files can be deleted while still being written to, so they use diskspace while their creating process is running, but not have a filename.

This makes it unfindable with the usual tools - you can use lsof to investigate which processes have open files.

Answer 7

If you can run software on the system, then xdiskusage will graphically show you which directories/files are eating your space. Extremely useful.

I believe KDE contains something similar.

If it's text-only and you cannot install extra software, then creative use of du will probably get you there as well.

Answer 8

here's something I cobbled together to track down some rogue processes on our database servers: rabbitfinder

#!/bin/sh
tree -s -f > /tmp/out1 && sleep 5 && tree -s -f > /tmp/out2; diff /tmp/out1 /tmp/out2 | egrep "\|--" | awk -F[ '{print $2}' | awk -F] '{print $2 }' | sort | uniq | xargs fuser -f | xargs ps -lFp

it's kinda kludgey and not very robust, but it works thusly:

1. generate a recursive tree list of the current directory
2. wait 5 seconds
3. generate another list
4. compare the two outputs
5. fuser the files that have changed size and

6. ps -lFp the files to show what process owns them

   user@poseidon:~$ tree -s -f > /tmp/out1 && sleep 5 && tree -s -f > /tmp/out2; diff /tmp/out1 /tmp/out2 | egrep "\|--" | awk -F[ '{print $2}' | awk -F] '{print $2 }' | sort | uniq | xargs fuser -f | xargs ps -lFp ./tmp/output:
   F S UID PID PPID C PRI NI ADDR SZ WCHAN RSS PSR STIME TTY TIME CMD 0 R 1000 14310 14275 23 80 0 - 1072 - 748 1 22:19 pts/2 00:00:06 dd if /dev/zero of ./output bs 1024 count 10000000