weave launch failure reported incorrectly

[rade]

# weave launch -foo
Failure during network configuration for container 3b6d181fc93f96207976e102311ea49daf381d3927f06b0904c68cc31a78826b:
Cannot open network namespace "10107": No such file or directory

This should have reported "Container 3b6d181fc93f96207976e102311ea49daf381d3927f06b0904c68cc31a78826b died". And I am quite sure it used to. I even get the same problem when running with --local. I wonder whether something has changed in docker that delays the updating of the container's metadata on death, because inserting a 'sleep 1' just before we run docker inspect makes the issue go away.

[rade]

I guess we could dispense with docker inspect and directly check whether the $CONTAINER_PID is dead (kill -0 $CONTAINER_PID).

However, having just tried that, I still get the same result. Which is odd.

[bboreham]

It seems to me there has always been a race between the weave script doing its check on why network setup has failed and Docker finding out that the process has died. Your test with kill -0 runs into the same problem: until the parent has waited on the dead process it still exists as a zombie.

[rade]

I must have been imagining things... The kill -0 works for me all the time now. Whereas the original docker inspect consistently fails to detect the dead weave in time.