Yeah, using --cgroups_enable_cfs. Sending parameters through DockerInfo is also an option, though I suspect the same behavior will result whenever systemd reconciles the state of the cgroup hierarchies with the state it constructs from unit files, since Docker also uses the fs interface to cgroups behind systemd's back. I guess this isn't really Docker specific, since the same behavior will likely result with the Mesos containerizer, but I was wondering if anyone else had encountered it or worked around it.
On Sat, Apr 16, 2016 at 6:37 AM, haosdent <[email protected]> wrote: > I am a bit curious about how you use cfs_quota with DockerContainerizer. Do > you use it via the Mesos Agent flag "--cgroups_enable_cfs"? Actually there > is another way to run docker with cfs via set `parameters` in `DockerInfo` > if your docker version greater than 1.7.0. > > The associated docker parameters are `cpu-period` and `cpu-quota`. > > On Sat, Apr 16, 2016 at 6:00 AM, Michael Browning <[email protected]> > wrote: > > > Hi all, > > > > We've run into an interesting issue with Docker and systemd that's > > hampering our attempts to use cfs_quota for CPU isolation with Docker > > containers. The issue is that Docker doesn't tell systemd when it > > changes cpu.cfs_quota_us, so when our Puppet run triggers a `systemctl > > daemon-reload` invocation, the quota is reset to -1. > > > > I assume this could be solved with the `Delegate=true` in the dockerd > unit > > file, but unfortunately we're not running a recent enough version of > > systemd to take advantage. Has anyone else ran into this when using > > DockerContainerizer with cfs_quota, or found a workaround? > > > > Regards, > > Michael > > > > > > -- > Best Regards, > Haosdent Huang >
