Seems like the failure is still happening:

https://stainless.corp.google.com/search?view=list&first_date=2018-11-03&last_date=2018-11-30&test=%5Enetwork_WiFi_DisconnectReason%5C.&status=FAIL&reason=DisconnectReason+not+logged&exclude_cts=true&exclude_not_run=false&exclude_non_release=true&exclude_au=true&exclude_acts=true&exclude_retried=true&exclude_non_production=false


It looks like network_WiFi_DisconnectReason simply searches for the supplicant DisconnectReason in net.log.  When I looked into those net.log collected by autotests, they are trimmed due to oversize and don't cover the period of time when the test ran. var/log_diff/net.log is also missing.  I'm wondering if it's something to do with log rotation.

We could probably do a few things to narrow down the issue:
- Before starting the test, force rotate net.log
- Update network_WiFi_DisconnectReason to also monitor D-Bus property update of supplicant

Ugh, I hate tests that try to parse logs like that. We unfortunately have quite a bit where that's the only option, but it'd be great if we could get this kind of stuff from D-Bus instead.

To confirm that shill receives the property update from supplicant and also make it observable, we could consider exposing a LastDisconnectReason property through shill. Then we don't need to rely on parsing net.log, which I also agree that it is pretty fragile.