[proxy] intermittent failure in entry point test

[rade]

I'm seeing the following every now and then...

$ ./620_proxy_entrypoint_command_test.sh 
Proxy uses correct entrypoint and command with weavewait
...
test #3 "proxy docker_on 192.168.48.11 run -e 'WEAVE_CIDR=10.2.1.1/24' --entrypoint='grep' false ^1$ /sys/class/net/ethwe/carrier" failed:
    program terminated with code 1 instead of 0

Running the failing command repeatedly on the host produces a failure about half the time...

root@host1:~# docker -H localhost:12375 run -e 'WEAVE_CIDR=10.2.1.1/24' --entrypoint='grep' false ^1$ /sys/class/net/ethwe/carrier
1
root@host1:~# docker -H localhost:12375 run -e 'WEAVE_CIDR=10.2.1.1/24' --entrypoint='grep' false ^1$ /sys/class/net/ethwe/carrier
1
FATA[0000] Error response from daemon: Container bf6d0a5ae5735d4443613cc165e3e4ac4e62639e720120c98aead839db59092d died 
root@host1:~# docker -H localhost:12375 run -e 'WEAVE_CIDR=10.2.1.1/24' --entrypoint='grep' false ^1$ /sys/class/net/ethwe/carrier
1
FATA[0000] Error response from daemon: Container 885c2204b90952cc0274c196b63d5b890cc84750ff0934bf67153280e7a0be3f died 

The log of the failed container contains 1.

[rade]

I've also had test#2 and test#5 fail.

[paulbellamy]

weave attach adds the network interface, which causes weavewait to run the main command. Then, weave attach carries on doing network-stuff. If the container exits before weave attach is finished, then weave attach throws the Container <...> died error. Making the weavewait interval longer causes the race condition to be less likely, but doesn't fix it.

Options:

1. Special-case and ignore that error (bleh)
2. Make weave attach atomic, from container's perspective (probably not possible...)
3. Change weavewait to block on USR2 signal, not the interface (IMO, the nicest option)

[paulbellamy]

So, we can't make weave attach "atomic" because the route needs to be setup after the interface is up, and the arp update needs to use the interface to ping our neighbours.