Solution & Estimations:
Current solution is to 1.) let Swarm launch tasks by Mesos 2.) for the other API, let Swarm send request to docker engine directly (red arrow)
Tasks of the solution (Mesos):
####POST /commit (~3 man month)
The container ID was changed after commit; Mesos did not know the new container ID. One option is to trace docker container in Mesos by container’s name (UUID).
####POST /containers/copy (Effort: ??)
Not sure how to copy between remote container.
####POST /containers/kill & stop (Effort: n/a)
If send this to Docker Engine directly, Mesos will mark it as TASK_FAILED instead of TASK_KILLED. Only support kill, the container will be stopped after killed.
####POST /containers/start & create (Effort: n/a)
Currently, Mesos launch tasks by “docker run” command (sub-process); so can not support those two API separately.
####POST /containers/exec & start (~1.5 man month)
Currently, Mesos did not support launch tasks by “docker exec”. Mesos need to support “docker exec”. But still can not support those two APIs separately.
Mesos:
###CLI parameters:
- Docker images
- network type (host, bridge)
- port mapping
- privileged
- parameters (including —label=xxx=yyy)
- force_pull_image
###Docker CLIs used by Mesos:
- docker run (launch task)
- docker version (version verify in Mesos)
- docker stop (kill/stop container)
- docker inspect (update status after the docker started)
- docker pull (get docker image)
- docker ps (reconcile docker status when slave recover)
- docker rm (remove docker images when executor exit)
comments powered by