cpuset: don't set cpuset.mems in the guest

[egernst]

Kata doesn't map any numa topologies in the guest. Let's make sure we
clear the Cpuset fields before passing container updates to the
guest. Without this, we could encounter a runtime failure:

process_linux.go:297: applying cgroup configuration for process caused
"failed to write 0,1 to cpuset.mems: writea 0,1 to cpuset.mems: write
0,1 to cpuset.mems: write /sys/fs/cgroup/cpuset .... /cpuset.mems:
numerical result out of range"": unknown

Note, in the future we may want to have a vCPU to guest CPU mapping and
still include the cpuset.Cpus. Until we have this support, clear this as
well.

Fixes: #2176
Depends-on: github.com/kata-containers/tests#2846

[egernst]

/test

[egernst]

/test

[codecov]

Codecov Report

Merging #2944 into master will increase coverage by 0.02%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #2944      +/-   ##
==========================================
+ Coverage   51.65%   51.67%   +0.02%     
==========================================
  Files         118      118              
  Lines       17430    17438       +8     
==========================================
+ Hits         9003     9011       +8     
  Misses       7342     7342              
  Partials     1085     1085              

[fidencio]

@egernst, the errors seem related to your change, please, take a look at them.

[egernst]

/test

[egernst]

@egernst, the errors seem related to your change, please, take a look at them.

Yeah, needed to update the tests. The existing cpuset tests assumed we were setting them in the guest. We shouldn't, imo - this is confusing, wrong, prone to not doing what you think it should, and doesn't add anything good at this point. :) Once we map vCPU - > guest CPU, we'll be able to apply a more appropriate CPUSet mapping from host to guest. Until then......

[egernst]

flakey test from centos: /cc @chavafg @GabyCT

1..1
not ok 1 test netmon
# (in test file integration/netmon/netmon_test.bats, line 60)
#   `[ "$after_interfaces" -gt "$before_interfaces" ]' failed
# 
# cb3d5ab7a2d048dd66846b3ab119b1da0e0c2970abcbe5636967416090988d20
# ba034ab9d2b94c43a80356291de15b1fd3815494643d34fd706752a435c72c5e
# containerA
# containerA
# test

[egernst]

/test

[egernst]

It seems the metrics job has been failing for a while now. Footprint has really crept up, I think we should look at 'why'. The particular test has failed since we've updated virtiofsd?

| *F* | memory-footprint     | 95.0% | 106.1% | 105.0% | 10.0% | 106.1% | 106.1% | 0.0%  | 0.0% |   1 |

[egernst]

/test

[jodh-intel]

Related:

• Memory footprint has been increased #2942
• metrics: Increase maxpercent for memory footprint limits tests#2840

[jodh-intel]

Thanks @egernst.

lgtm, but yeah, would be good to add a UT to this PR 😄

Also, presumably this needs porting to 2.0?

[fidencio]

lgtm!

I'd like to have @devimc's input here before getting this one merged!

[devimc]

@egernst thanks, could you update the limitations document?

[fidencio]

I've added the "backport" label as I'd really like to see this one as part of the stable-1.11 branch.

[egernst]

How should we handle metrics CI? Blocking until the test is fixed?

[fidencio]

@jodh-intel, side note, I've changed your labels.
Seems that "forward-port" and "backport" labels are referring to the specific PR, whether this specific PR is a forward-port or a backport of a patch.

The labels we should be using, though, are the "needs-backport" and "needs-forward-port"

[jodh-intel]

I've added the "backport" label as I'd really like to see this one as part of the stable-1.11 branch.

@fidencio - Agreed. However, the label to use for that is needs-backport: the backport label is for a PR that itself is a backport PR. The behaviour is summarised on kata-containers/kata-containers#634 (comment), but I'll be updating the docs hopefully today on kata-containers/community#172 and letting the ML know...

[jodh-intel]

@fidencio - \o/ :smile:

[fidencio]

/test

[jodh-intel]

@egernst - I've respun the failing CI's. Are you planning on working on a 2.0 port of this?

[jodh-intel]

forward port PR: kata-containers/kata-containers#933