launchpad.net/juju-core/juju tests sometimes fail mgo tip

[Michael Hudson-Doyle]

I've seen the

Waiting for sockets to die: 0 in use, 1 alive
/home/mwhudson/goplay/src/launchpad.net/juju-core/testing/mgo.go:333:
    c.Fatal("Test left sockets in a dirty state")

thing in other failures.

[Andrew Wilkins (axwalk)]

This is not particular to gccgo - there's a race condition somewhere, causing this even with gc.

[John A Meinel (jameinel)]

Last time we ran into a failure like this there was a race in the mgo driver.
I also seem to remember that we had to use a fixed version of mgo (241?) because of other incompatibilities. I wonder if we got a newer version and it brought in another race.

[John A Meinel (jameinel)]

As noted, this isn't gccgo specific.

[Michael Hudson-Doyle (mwhudson)]

Ah well I just ran "go get juju-core/..." so I probably did get mgo tip?
And in fact you need a very recent fix for it to work at all with gccgo.
On 21 Nov 2013 01:50, "John A Meinel" <email address hidden> wrote:

> As noted, this isn't gccgo specific.
>
> --
> You received this bug notification because you are subscribed to the bug
> report.
> https://bugs.launchpad.net/bugs/1252859
>
> Title:
> launchpad.net/juju-core/juju tests fail with gccgo
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/juju-core/+bug/1252859/+subscriptions
>

[Dave Cheney (dave-cheney)]

oop, I think you are right, somewhere along the way we did bump the
version of the mongo driver.

On Thu, Nov 21, 2013 at 6:32 AM, Michael Hudson-Doyle
<email address hidden> wrote:
> Ah well I just ran "go get juju-core/..." so I probably did get mgo tip?
> And in fact you need a very recent fix for it to work at all with gccgo.
> On 21 Nov 2013 01:50, "John A Meinel" <email address hidden> wrote:
>
>> As noted, this isn't gccgo specific.
>>
>> --
>> You received this bug notification because you are subscribed to the bug
>> report.
>> https://bugs.launchpad.net/bugs/1252859
>>
>> Title:
>> launchpad.net/juju-core/juju tests fail with gccgo
>>
>> To manage notifications about this bug go to:
>> https://bugs.launchpad.net/juju-core/+bug/1252859/+subscriptions
>>
>
> --
> You received this bug notification because you are subscribed to juju-
> core.
> Matching subscriptions: MOAR JUJU SPAM!
> https://bugs.launchpad.net/bugs/1252859
>
> Title:
> launchpad.net/juju-core/juju tests fail with gccgo
>
> Status in juju-core:
> Triaged
>
> Bug description:
> I've seen the
>
> Waiting for sockets to die: 0 in use, 1 alive
> /home/mwhudson/goplay/src/launchpad.net/juju-core/testing/mgo.go:333:
> c.Fatal("Test left sockets in a dirty state")
>
> thing in other failures.
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/juju-core/+bug/1252859/+subscriptions

[Michael Hudson-Doyle (mwhudson)]

And this week, it seems, the tests pass with gccgo. So yeah, this isn't a real gccgo failure...

[Ian Booth (wallyworld)]

We bunch of mgo related test fixes have recently landed in trunk. Closing this, but re-open if issue still occurs.