-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/45454/
-----------------------------------------------------------
Review request for mesos, Jie Yu, Joris Van Remoortere, and Timothy Chen.
Bugs: MESOS-3573
https://issues.apache.org/jira/browse/MESOS-3573
Repository: mesos
Description
-------
This change modifies the docker containerizer to cleanup docker
containers left from another agent instance. The containers can
become orphans due to any of the scenarios mentioned here:
http://bit.ly/1RxCpPl
This change modifies the logic to invoke docker `ps` on all
containers on the agent instead of limiting itself to the
current slaveID. This change also means that running multiple
agent instances on the same host might not work well for docker
containers from now on i.e. another agent instance might
cleanup the docker containers that belong to another instance.
The cgroup isolators/linux launcher for the Mesos containerizer
anyways don't recommend running multiple instances of the agent
on the same host.
In case one still wants to run multiple agent instances on a
test cluster using the docker containerizer, we can use the
`--no-docker_kill_orphans` flag and then kill the docker
containers manually using a script.
Diffs
-----
src/slave/containerizer/docker.hpp 89d450e10a84f24ddd46d517e2b4b46ab02c4fda
src/slave/containerizer/docker.cpp c5007a311ae9c1766dd4522ccbddbdb506d4ae4e
Diff: https://reviews.apache.org/r/45454/diff/
Testing
-------
make check (Test is added as part of the next review in the chain)
Would follow up with an email on user@ to notify them about the potential
gotchas with running multiple agent instances in production.
Thanks,
Anand Mazumdar
