| 2018-03-12 11:10:04 |
John A Meinel |
description |
As seen here:
http://10.125.0.203:8080/job/RunUnittests-race-amd64/176/console
[LOG] 0:05.024 DEBUG juju.api error dialing websocket: dial tcp 127.0.0.1:37180: context deadline exceeded
apiclient_test.go:797:
c.Assert(err, gc.ErrorMatches, `cannot log in: context deadline exceeded`)
... error string = "unable to connect to API: dial tcp 127.0.0.1:37180: context deadline exceeded"
... regex string = "cannot log in: context deadline exceeded"
Oddly, we see this before the test info:
=== RUN TestAll
2018/03/12 11:00:27 http: TLS handshake error from 127.0.0.1:48562: EOF
I wonder if the context deadline we are setting is sometimes so short that the TLS handshake fails first, and then we get cut off before we even get to Log In which is why we get a failure to "connect to API" rather than a failure "cannot log in".
I'll check if the code actually cares, or if we can just look for either message.
Note, this *particular* instance of the failure is under "-race" but probably that just changes the timing to make it more possible. |
As seen here:
http://10.125.0.203:8080/job/RunUnittests-race-amd64/176/console
github.com/juju/juju/api
FAIL: apiclient_test.go:770: apiclientSuite.TestOpenTimesOutOnLogin
[LOG] 0:05.024 DEBUG juju.api error dialing websocket: dial tcp 127.0.0.1:37180: context deadline exceeded
apiclient_test.go:797:
c.Assert(err, gc.ErrorMatches, `cannot log in: context deadline exceeded`)
... error string = "unable to connect to API: dial tcp 127.0.0.1:37180: context deadline exceeded"
... regex string = "cannot log in: context deadline exceeded"
Oddly, we see this before the test info:
=== RUN TestAll
2018/03/12 11:00:27 http: TLS handshake error from 127.0.0.1:48562: EOF
I wonder if the context deadline we are setting is sometimes so short that the TLS handshake fails first, and then we get cut off before we even get to Log In which is why we get a failure to "connect to API" rather than a failure "cannot log in".
I'll check if the code actually cares, or if we can just look for either message.
Note, this *particular* instance of the failure is under "-race" but probably that just changes the timing to make it more possible. |
|
