Deploying to maas ppc64le with proxies kills websocket

[Marco Ceppi]

Juju Version: 1.25.0

The allwatcher/websocket suddenly closes after a certain time period when watching an environment on a maas machine which requires proxies for some but not all web requests.

console output: http://juju-ci.vapour.ws:8080/job/charm-bundle-test-power8/10/console
all-machines log: http://juju-ci.vapour.ws:8080/job/charm-bundle-test-power8/10/artifact/all-machines.log

maas server and all internal maas ips have been explicitly added to the no-proxy environment variable

[Cheryl Jennings (cherylj)]

In all-machines.log, I don't see that any proxy information is set:

machine-0: 2015-11-10 15:57:46 DEBUG juju.worker.proxyupdater proxyupdater.go:151 new proxy settings proxy.Settings{Http:"", Https:"", Ftp:"", NoProxy:""}
machine-0: 2015-11-10 15:57:47 DEBUG juju.worker.proxyupdater proxyupdater.go:170 new apt proxy settings proxy.Settings{Http:"", Https:"", Ftp:"", NoProxy:""}

There may be pieces missing that are in machine-0.log. Could you attach that file?

[Eric Snow (ericsnowcurrently)]

FWIW, I ran the test again and it failed in the same way but in a different spot:

  http://juju-ci.vapour.ws:8080/job/charm-bundle-test-power8/55/console

[Eric Snow (ericsnowcurrently)]

Note that the all-machines.log (which Marco uploaded) appears to be about 33 minutes behind the console log.

[Eric Snow (ericsnowcurrently)]

I haven't made a lot of headway on this. Thus far it isn't clear to me where the problem is, particularly since it's failing in different places across multiple requests.

[Marco Ceppi (marcoceppi)]

We think this may be a networking issue between where the test runner is and the environment. We're working how to replicate this directly on the POWER8 machine to remove this as a potential bottleneck.