[BEAM-6020] Rework Docker-inside-Docker for JobServer container

• Type: Improvement
• Status: Open
• Priority: P3
• Resolution: Unresolved
• Component/s: runner-flink
• Labels: portability-flink
• Assignee: Unassigned
• Reporter: Maximilian Michels
• Created: 08/Nov/18 10:57
• Updated: 03/Jun/22 22:53

Description

Docker-inside-Docker causes problems with permissions. For example:

Caused by: java.lang.Exception: The user defined 'open()' method caused an exception: java.io.IOException: Cannot run program "docker": error=13, Permission denied
	at org.apache.flink.runtime.operators.BatchTask.run(BatchTask.java:498)
	at org.apache.flink.runtime.operators.BatchTask.invoke(BatchTask.java:368)
	at org.apache.flink.runtime.taskmanager.Task.run(Task.java:712)
	... 1 more
Caused by: org.apache.beam.repackaged.beam_runners_java_fn_execution.com.google.common.util.concurrent.UncheckedExecutionException: java.io.IOException: Cannot run program "docker": error=13, Permission denied
	at org.apache.beam.repackaged.beam_runners_java_fn_execution.com.google.common.cache.LocalCache$LocalLoadingCache.getUnchecked(LocalCache.java:4994)

	... 7 more

See discussion:
https://lists.apache.org/thread.html/b2b97efaf1f4a6e283eaf92e05605f10f405e9eb9a9c41896041d0ac@%3Cdev.beam.apache.org%3E

We could get rid of Docker-inside-Docker entirely. It would be cleaner to start the SDK harness container alongside with JobServer container.

Alternatively, it should be possible to use an image with Docker pre-installed, which should fix the permissions errors which come from mounting the Docker executable of the host.