[MESOS-5927] Unable to run "scratch" Dockerfiles with Unified Containerizer.

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.0.0
• Fix Version/s: 0.28.3, 1.0.1
• Component/s: containerization
• Labels: containerizer filesystem mesosphere
• Epic Link: Unified Container
• Sprint: Mesosphere Sprint 40
• Story Points: 3
• Assignee: Gilbert Song
• Reporter: Philip Winder
• Shepherd: Jie Yu
• Created: 29/Jul/16 07:34
• Updated: 03/Aug/16 00:06
• Resolved: 03/Aug/16 00:06

Description

It is not possible to run Docker containers that are based upon the "scratch" container.

Setup: Mesos 1.0.0 with the following Mesos settings:

echo 'docker' | sudo tee /etc/mesos-slave/image_providers
echo 'filesystem/linux,docker/runtime' | sudo tee /etc/mesos-slave/isolation

Recreate: From a Master or Slave, run:

mesos-execute --command='echo ok' --docker_image=hello-seattle --master=localhost:5050 --name=test

Effect: The container will crash with messages from Mesos reporting it can't mount folder x/y/z. E.g. can't mount /tmp. This means you can't run any container that is not a "fat" container (i.e. one with a full OS). E.g. error:

Failed to enter chroot '/var/lib/mesos/provisioner/containers/fed6add8-0126-40e6-ae81-5859a0c1a2d4/backends/copy/rootfses/4feefc8b-fd5a-4835-95db-165e675f11cd': /tmp in chroot does not existI0729 07:49:56.753474 4362 exec.cpp:413] Executor asked to shutdown

Expected: Run without issues.

Use case: We use scratch based containers with static binaries to keep the image size down. This is a common practice.