[ https://issues.apache.org/jira/browse/MESOS-2595?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14545685#comment-14545685 ]
Jay Buffington commented on MESOS-2595: --------------------------------------- [~tnachen] Is the new docker executor always pid 1 inside the container when the scheduler doesn't use ExecutorInfo? Does it do proper pid 1 things like reap orphans? > Create docker executor > ---------------------- > > Key: MESOS-2595 > URL: https://issues.apache.org/jira/browse/MESOS-2595 > Project: Mesos > Issue Type: Improvement > Components: docker > Reporter: Timothy Chen > Assignee: Timothy Chen > > Currently we're reusing the command executor to wait on the progress of the > docker executor, but has the following drawback: > - We need to launch a seperate docker log process just to forward logs, where > we can just simply reattach stdout/stderr if we create a specific executor > for docker > - In general, Mesos slave is assuming that the executor is the one starting > the actual task. But the current docker containerizer, the containerizer is > actually starting the docker container first then launches the command > executor to wait on it. This can cause problems if the container failed > before the command executor was able to launch, as slave will try to update > the limits of the containerizer on executor registration but then the docker > containerizer will fail to do so since the container failed. > Overall it's much simpler to tie the container lifecycle with the executor > and simplfies logic and log management. -- This message was sent by Atlassian JIRA (v6.3.4#6332)