Skip to content

MeetingMinutes: 2015 08 26

Vincent Batts edited this page Sep 4, 2015 · 3 revisions

HomeMeetingMinutes:

Discussion w/ Video:

Video at https://bluejeans.com/1771332256/

Topics:

Attendees:

All attendees below participated in the BlueJeans video

  • Abhijeeth Nuthan (Abhi)
  • Brandon Philips
  • crosbymichael
  • Daniel Dao
  • Jesse Butler (jlb13)
  • Julz Friedman
  • Liangchenye
  • lk4d4
  • Mrunal Patel
  • Phil Estes
  • Timothy Chen
  • Trevor (wking)
  • Vincent Batts
  • Vishnu Kannan
  • Rob Dolin
  • Ross Gardler
  • Tianon
  • Will Pragnell

#Minutes:

  • https://github.com/opencontainers/specs/pull/87
    • mrunal / crosbymichael
    • need feedback from CRIU folks. crosbymichael to follow-up
    • why do we need the cgroup, namespaces stuff (philips)
    • the constraint of //state.json and the id value in the json
    • Motivation for:
      • id: so teardown hooks can do something with network teardown?
      • pid: this gives you access to cgroups, namespaces, open files, etc.
      • root: so state-consumers can mount things into that filesystem without entering the container’s mount namespace.
    • use case of /run/oci being a tmpfs (philips)
      • cadvisor to discover the containers for reporting on (crosbymichael)
      • like machinectl to discover containers
      • exec would need it to run pid in existing containers
      • perhaps env variable to set this root path
      • Needs an OCI version field in the state file (in the file for hooks to use too). More version discussion in #110. I think we want the same version semantics for the spec and state files.
    • michael to make updates
    • Hmm, if runV and runC are sharing the same state, maybe the PID isn’t sufficient for looking up container information?
  • https://github.com/opencontainers/specs/pull/88
    • lock and chown a bundle for userns mapping
    • it would be ideal to have the kernel vfs support a uid/gid shift, so we do not be concerned with chown
    • everyone give it a once over, so we can merge
    • drop rootfs requirement, since folks could have valid use-cases with external-to-bundle root directories.
  • https://github.com/opencontainers/specs/issues/114
    • simple way to begin handling performance and memory pressures
    • dictating cgroup hierarchies
    • we want to handle with this, multiple processes in a shared life-cycle. Like side-car containers, live and debug versions of an application

Next Meeting:

Below are some proposed agenda items for the next meeting: