Travis-CI nearly unusable due to extremely long queuing.

[daspecster]

For gcloud-python and other languages, Travis-CI has been taking many hours to run builds.

We should address this in order to maintain gcloud-python quality and reliability.

[theacodes]

This is the case across the GoogleCloudPlatform org.

[daspecster]

@jonparrott, hopefully things can get sorted out in the GoogleCloudPlatform travis account. If not I guess we'll need to get a secondary plan together.

[waprin]

Some options our team is considering:

1. Figure out if Travis can fix it

2. I have an internal Jenkins instance. It's firewall off but it can upload build output and build history to GCS buckets as comments on the PRs. Take a look at github.com/kubernetes/minikube PRs to see what it could look like. If you're interested happy to set it up but I'm on vacation next 2 weeks though I'm sure Jon could figure it out too.

3. CircleCI

4. Turn your test into a Docker build and then use Cloud Container Builder.

[daspecster]

@waprin, 2 & 4 sound really cool. I think having the badge for a public project is valuable though.
I'm not sure what other people think?

I did notice that sometime this weekend, things with travis have seemed to speed up. If this holds out for a few days(September 1st) then I'll close this.

[tseaver]

I wonder if the huge backlog was due to Travis OSX problem.

[daspecster]

I really hope they don't have an architecture that is tightly coupled enough to have one platform cause issues with other platforms.
I guess if they had a bunch of backed up OSX builds and needed to swing vm resources over to OSX to catch back up?

[dhermes]

Closing this out because Travis seems to have sped back up. Maybe not though? Feel free to re-open.