Skip to content

Regression in top: Total CPU usage is always 0 (zero) while using a core limitation by cgroup-cpuset

Guido Jäkel requested to merge gjaekel/procps:patch-1 into master

I'm using Gentoo for LXC host and containers. After upgrading the procps-ng package from 3.3.12 to 3.3.15, in the top application the value of summed CPU usage inside a container is always 0 (zero). We're using a pool of enterprise blade servers (2 CPU NUMA -> lots of cores).

  • This issue shows up in the containers only, but not on the host. There core reason seems that the cgroup-cpuset for my containers is always limited to a subset of cores. If I remove the limit and assign all cores to a container, the issue vanish.
  • If I switch the display to show the value for individual CPUs, this display "works"
  • This was not the case with the former used 3.3.12

A quick look at the source code of top show that it seems to use /proc/stat. Watching a 'head -1 /proc/stat' show changing values for the cpu line - no surprise since the v3.12 of top will work.

Then I compare the sources of 3.3.12 and 3.3.15, I may found a regression: The statement in the patch disappeared and if I add it again, the CPU sum display seems to work, again.

Therefore: Might this be a regression? Or was this deletion intended and something other is broken?

with greetings

Guido

Signed-off-by: Guido Jäkel G.Jaekel@DNB.DE

Edited by Guido Jäkel

Merge request reports