using lxc_conf in start

[ureyes84]

How do I use lxc_conf in start?

I'd like to set cpu and cpu shares, something like this:

docker run -i -t -lxc-conf="lxc.cgroup.cpu.shares=1" -lxc-conf="lxc.cgroup.cpuset.cpus=0" my_image /bin/bash stress_cpu.sh

According to the documentation I should be able to pass a dictionary:

lxc_conf allows to pass LXC configuration options in dict form.

So my code looks like this:

self.client.start(container, lxc_conf = self.get_lxc_conf())
def get_lxc_conf(self):        
  return {
  #Restrict container to a core
  'lxc.cgroup.cpuset.cpus' : str(self.cpu_number), 
  #All containers get the same share
  'lxc.cgroup.cpu.cpu.shares' : str(1) 
}

However, I can't get it work, I keep getting the following error message:

500 Server Error: Internal Server Error ("json: cannot unmarshal object into Go value of type []docker.KeyValuePair")

I get the same error even with the dictionary example in the API docs:

def get_lxc_conf(self):
    return {"lxc.utsname":"docker"}

I've tried passing lists and strings

Thanks!

[ureyes84]

This doesn't seem to error out but I it doesn't work either (takes all CPUs):

self.client.start(container, lxc_conf = self.get_lxc_conf())
def get_lxc_conf(self):
return[
    #All containers get the same share
    {'lxc.cgroup.cpu.shares':1},
    #Restrict container to a core
    {'lxc.cgroup.cpuset.cpus':self.cpu_number}
]

The CLI counterpart is able to limit this CPU-intensive script to a single CPU:

docker run -i -t -lxc-conf="lxc.cgroup.cpu.shares=1" -lxc-conf="lxc.cgroup.cpuset.cpus=0" my_image /bin/bash stress_cpu.sh

[ureyes84]

Interestingly, after running

docker run -i -t -lxc-conf="lxc.cgroup.cpu.shares=1" -lxc-conf="lxc.cgroup.cpuset.cpus=0" my_image /bin/bash stress_cpu.sh

I can just do

docker start 104059c2e613

And the previous resource limits are applied. However if I do afterwards the equivalent (I think) with docker-py, the resource limits are ignored:

self.docker_client.start('104059c2e613')

[denibertovic]

@ureyes84 When using the docker CLI the run command has a -c flag that's used for setting CPU shares.

As for docker-py. the lxc_conf param in the Client.start method expects a list of dicts like so:

[{"Key":"lxc.cgroup.cpu.shares","Value":"1"}, {"Key": ..., "Value": }]

And, both methods (CLI and docker-py) set that correctly in the lxc config (you can verify this in /var/lib/docker/containers/{container_id}/config.

You can test with:

client.create_container('ubuntu', '/bin/bash -c "dd if=/dev/zero of=/dev/null | dd if=/dev/zero of=/dev/null | dd if=/dev/zero of=/dev/null | dd if=/dev/zero of=/dev/null"')

and then:

c.start('04095d4fc948', lxc_conf=[{'Key': 'lxc.cgroup.cpuset.cpus', 'Value': '1'}])

Note that the restriction we use to restrict the number of cores to be used is 'lxc.cgroup.cpuset.cpus'
and not lxc.cgroup.cpu.shares (this one does something different and depends on the current usage of the CPU, ie.
if nobody is using their shares the one container will get all the cores to it's disposal until some other container
needs it's shares.).

I hope this helps.

[ureyes84]

Thank you very much! They Key Value syntax helped. I didn't see that coming though, do you think the docs should be updated?

Also, thank you for explaining the cpu shares. My intent is to limit resources, I think restricting the number of cores is the way to go.

Quick question regarding this, say I have a dedicated server that creates ALL containers with:

[{"Key":"lxc.cgroup.cpu.shares","Value":"1"}]

Does it really make a difference adding this parameter? I was wondering whether this is useful only when you need to give more "weight" to a specific container and set the Value parameter to something higher.

[denibertovic]

You are right about the docs. And tnx for the pull request, I've commented there.

Regarding cpu.shared, you are right, it's suited for giving priority to certain containers by setting a higher value.

[denibertovic]

I think we can close this.

[ureyes84]

Thanks again for your help