[MESOS-6327] Large docker images causes container launch failures: Too many levels of symbolic links.

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 1.0.0, 1.0.1
• Fix Version/s: 1.1.2, 1.2.1, 1.3.0
• Component/s: containerization, docker

Environment

centos 7.2 (1511), ubuntu 14.04 (trusty). Replicated in the Apache Aurora vagrant image

• Assignee: Chun-Hung Hsiao
• Reporter: Rogier Dikkes
• Shepherd: Gilbert Song
• Created: 07/Oct/16 08:58
• Updated: 18/Apr/17 06:36
• Resolved: 18/Apr/17 06:36

Description

When deploying Mesos containers with large (6G+, 60+ layers) Docker images the task crashes with the error:

Mesos agent logs:
E1007 08:40:12.954227 8117 slave.cpp:3976] Container 'a1d759ae-5bc6-4c4e-ac03-717fbb8e5da4' for executor 'thermos-www-data-devel-hello_docker_image-0-d42d2af6-6b44-4b2b-be95-e1ba93a6b365' of framework df
c91a86-84b9-4539-a7be-4ace7b7b44a1-0000 failed to start: Collect failed: Collect failed: Failed to copy layer: cp: cannot stat ‘/var/lib/mesos/provisioner/containers/a1d759ae-5bc6-4c4e-ac03-717fbb8e5da4/b
ackends/copy/rootfses/5f328f72-25d4-4a26-ac83-8d30bbc44e97/usr/share/zoneinfo/right/Asia/Urumqi’: Too many levels of symbolic links
... (complete pastebin: http://pastebin.com/umZ4Q5d1 )

How to replicate:
Start the aurora vagrant image. Adjust the /etc/mesos-slave/executor_registration_timeout to 5 mins. Adjust the file /vagrant/examples/jobs/hello_docker_image.aurora to start a large Docker image instead of the example. (you can use anldisr/jupyter:0.4 i created as a test image, this is based upon the jupyter notebook stacks.). Create the job, watch it fail after x number of minutes.

The mesos sandbox is empty.

Aurora errors i see:
28 minutes ago - FAILED : Failed to launch container: Collect failed: Collect failed: Failed to copy layer: cp: cannot stat ‘/var/lib/mesos/provisioner/containers/93420a36-0e0c-4f04-b401-74c426c25686/backends/copy/rootfses/6e185a51-7174-4b0d-a305-42b634eb91bb/usr/share/zoneinfo/right/Asia/Urumqi’: Too many levels of symbolic links cp: cannot stat ...
Too many levels of symbolic links ; Container destroyed while provisioning images
(complete pastebin: http://pastebin.com/uecHYD5J )

To rule out the image i started this and more images as a normal Docker container. This works without issues.

Mesos flags related configured:
-appc_store_dir
/tmp/mesos/images/appc
-containerizers
docker,mesos
-executor_registration_timeout
5mins
-image_providers
appc,docker
-image_provisioner_backend
copy
-isolation
filesystem/linux,docker/runtime

Affected Mesos versions tested: 1.0.1 & 1.0.0

Issue Links

• is duplicated by: MESOS-6875 Copy backend fails to copy container
• relates to: MESOS-7280 Unified containerizer provisions docker image error with COPY backend