2

We have regularly high load average. With the help of a zabbix agent, I could narrow it down to be mainly because of cpu iowait (and lastly disk io). I am not allowed to install any additional packages on the server but have root rights and want to investigate the issue. By now I know the affected partitions.

There are no tools like iostat, iotop, sar, etc. available. So I was looking around if there is, like so often, a (pseudo-)file that contains the info that is read by those tools. Since we use RAID, I first looked into /proc/mdstat which gives the mapping of our devices /dev/mdX to the disk partitions. Then I looked into /proc/diskstat and with the help of https://www.kernel.org/doc/html/latest/admin-guide/iostats.html I could find the partitions that are affected by the most IO.

How can I nail it down to files or processes from here? Can lsof be helpful? This is available.

nox
  • 121
  • 3
  • Have a look into `/proc/PROCESS_ID/io`. This file contains the number of bytes read and written by the process. I understand that the bytes actually transferred to/from disk is not necessarily accurate, but the numbers should help. See the manual page for *proc*. – berndbausch Jan 13 '21 at 14:40
  • On this system, there is no `/proc/pid/io`, only `/proc/pid/fd`. This doesn't get me much further though. – nox Jan 13 '21 at 14:55
  • Have you considered just building the tools locally instead of installing the package? If the computer has the development tools needed it may be easier to bust build them yourself. – miguelbernadi Mar 26 '21 at 18:05
  • Thanks @miguelbernadi, I did consider it, but the necessary dev tools are not installed. Currently we reduced the load and the system runs fine which is why I didn't continue this too eagerly. – nox Mar 29 '21 at 09:23

0 Answers0