Scenario 3

Tutorial - Docker Images

Scenario 3: Docker Images


Start by deploying this dockerimage.json file:

$ dcos marathon app add

We see the app fail almost immediately:

Pic of failure


As we learned earlier, with application failures the first step is to check the task logs.

Pic of empty log output

Unfortunately, it is completely empty. Normally we would at least see some output from the setup of the task. This is especially peculiar behavior.

So [Step 2] is to check the scheduler logs — in this case Marathon:

$ dcos service log marathon

which should produce something like the following output in response:

Mar 27 21:21:11[5954]: [2018-03-27 21:21:11,297] INFO  Received status update for task docker-image.c4cdf565-3204-11e8-8a20-82358f3033d1: TASK_FAILED (

Mar 27 21:21:11[5954]: ') (mesosphere.marathon.MarathonScheduler:Thread-1723)

However, this does not shed much light on why the task failed. So then to Step 3 of our strategy: check the Mesos agent logs using:

$ dcos node log --mesos-id=$(dcos task docker-image  --json | jq -r '.[] | .slave_id') --lines=100

to output something resembling the following:

8-4520-af33-53cade35e8f9-0001 failed to start: Failed to run 'docker -H unix:///var/run/docker.sock pull noimage:idonotexist': exited with status 1; stderr='Error: image library/noimage:idonotexist not found

2018-03-27 21:27:15: '

2018-03-27 21:27:15: I0327 21:27:15.325984  4765 slave.cpp:6227] Executor 'docker-image.9dc468b5-3205-11e8-8a20-82358f3033d1' of framework 6512d7cc-b7f8-4520-af33-53cade35e8f9-0001 has terminated with unknown status

It looks like the specific Docker image could not be found, perhaps because it doesn’t exist. Does the image exist in the specified location (in this case noimage:idonotexist in Dockerhub)? If it does not, you will have to correct the location or move the file to the specified location. Furthermore, was there an error in the specified location or file name? Lastly, check whether the container image registry is accessible (especially when using a private registry). s

General Pattern

Being an application error, we again start by looking at task logs, followed by scheduler logs.

In this case we have a Docker daemon-specific issue. Many such issues can be uncovered by examining the Mesos Agent logs. In some cases, where we need to dig deeper, accessing the Docker daemon logs is required. First, ssh into the master node:

$ dcos node ssh --master-proxy --mesos-id=$(dcos task --all | grep docker-image | head -n1 | awk '{print $6}')

then to get the logs:

$ journalct1 -u docker

Please note the more complex pattern used here to retrieve the mesos-id in comparison to the earlier example. This pattern lists previously failed tasks as well as running tasks, whereas the earlier pattern only lists running tasks.



$ dcos marathon app remove docker-image