Project

General

Profile

action #35877

Updated by okurz about 6 years ago

## Motivation 
 Often we have the situation that tests fail because "something might be slow". But the question might be is it CPU or I/O? To answer this we should look for I/O load in post_fail_hooks 

 ## Acceptance criteria 
 * **AC1:** generic post_fail_hook in os-autoinst-distri-opensuse checks for I/O load on failures as one of the first steps 

 ## Suggestion 
 * `sed -n 's/^.*sda / /p' /proc/diskstats | cut -d' ' -f10` will show the value of "0" if there is no I/O pending and a higher number if there *is* I/O pending for the first disk. Use this or something comparable as one of the very first steps in a generic post_fail_hook, e.g. the one in lib/opensusebasetest.pm 


 ## Alternatives 

 * `sudo iotop --batch --iter=3 --processes --quiet --only` but it requires the package "iotop" installed which IMHO we do not even have available everywhere 


 ## References 

 * https://www.kernel.org/doc/Documentation/iostats.txt 
 *

Back