bug: memory.memsw.usage_in_bytes : no such file or directory

[ivan-kiselev]

Hi there!

I have a problems with Stat() function in cgroups package:

Code example(approximately):

control, err = cgroups.New(cgroups.V1, cgroups.StaticPath(cgroupName), &specs.LinuxResources {})
err := control.Add(cgroups.Process{Pid:int(pid)})
stats, err := control.Stat()
usage := stats.Cpu.Usage.Total
fmt.Printf("\n%v", usage)

It's just a part of a code, but its quite simple for seeing the logic I go through.

But when I try to run it:

• Process that I expect to be moved into cgroupName is moved, it's ok.
• My code falls into panic with /sys/fs/cgroup/memory/somecgroupname/memory.memsw.usage_in_bytes: no such file or directory

So, what have I found: memory.memsw.usage_in_bytes file isnt always present in system because its optionaly cgroups feature that could be activated by passing swapaccount=1 into kernel (actualy, I'm using Ubunt/Debian systems in my environment, so I have the same situation as I linked you to)

Additional links about kernel parameters that has impackt to file existing factor:

http://cateee.net/lkddb/web-lkddb/MEMCG_SWAP.html
http://cateee.net/lkddb/web-lkddb/MEMCG_SWAP_ENABLED.html.

So, it looks like we need some check about if the file exists or it doesnt to prevent that kind of panic.
May be I will try to figure it out by myself and create PR then, but if you have enough time and see quite easy way to do it, it would be nice.

Anyway, thank you for opensourcing that again!

[crosbymichael]

First is the panic coming from your code or this cgroups package? If its coming from the package can you post the stack and I can fix it or you can submit a PR if you have time.

To fix this fast you should be able to pass an error handler for the stats function. This is one already in this package to ignore isnotexist errors.

control.Stats(cgroups.IgnoreNotExists)

You could also make your own ErrorHander func and pass to that method if you want it to do other things.

[ivan-kiselev]

oh, control.Stat(cgroups.ErrorHandler(cgroups.IgnoreNotExist)) did the trick! Thx, and sorry for taking your time with my inattention: didnt open errors.go file before.

[ivan-kiselev]

P.S. May be it's worthy to be noticed in README.md?

[crosbymichael]

Ya, we can put an example in the readme

[point-free]

I'm running into the same issue. If you pass cgroups.ErrorHandler(cgroups.IgnoreNotExist), won't that prevent the rest of the memory stats from being read (e.g. kmem and kmem.tcp)? It seems like we'll return as soon as we hit a single error: https://github.com/containerd/cgroups/blob/master/memory.go#L163

[mariomac]

The changes introduced in this PR should help dealing with this issue: #98