Jay Buffington created MESOS-2278:
-------------------------------------
Summary: killing a task while containerizer is launching results
in "Unregistered Executor" core message
Key: MESOS-2278
URL: https://issues.apache.org/jira/browse/MESOS-2278
Project: Mesos
Issue Type: Story
Components: containerization
Reporter: Jay Buffington
I have only verified this bug with the docker containerizer, but I assume it
exists in all containerizers.
If the scheduler is using an executor and sends a killTask to mesos after the
slave has received the task, but before the task is in TASK_RUNNING state, the
scheduler will receive a statusUpdate with a message of "Unregistered Executor"
and this error appears in the mesos-slave log:
{noformat}
W0127 19:11:52.219424 28172 slave.cpp:1465] Killing the unregistered executor
'<executor id>' of framework <framework id> because it has no tasks
E0127 19:11:52.222455 28172 slave.cpp:2323] Failed to update resources for
container <container id> of executor <task id> running task <task id> on status
update for terminal task, destroying container: No container found
{noformat}
This happens often in the Aurora case:
- user does an "aurora create" of a docker job
- slave does "docker pull" which takes a while (~2-3 minutes)
- user does "aurora kill" while the slave is doing the "docker pull"
There should be no error generated in this case, because everything is working
properly. The status message the scheduler gets back should be "Instructed to
kill task"
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
