Fail fast for OSX CI runs.
Review Request #1894 - Created March 10, 2015 and submitted
The Rube Goldberg machine that enables OSX CI for pants has a systemic failure mode when using multiple shards. The sha the shards need to check out can be nuked by force pushes to the https://github.com/pantsbuild/pants-for-travis-osx-ci repo in the https://github.com/pantsbuild/pants CI job. This change accounts for this common and known failure mode by failing faster to try to up the green throughput during periods of high commit volume on https://github.com/pantsbuild/pants master. .travis.osx.yml | 10 ++++++++++ 1 file changed, 10 insertions(+)
.travis.osx.ymllinted green:jsirois@gill ~/dev/3rdparty/jsirois-pants4 (jsirois/osx-ci/fail-fast *) $ travis lint .travis.osx.yml Hooray, .travis.osx.yml looks valid :)
CI went green here:
I noticed this phenomenon looking at the build history here: https://travis-ci.org/pantsbuild/pants-for-travis-osx-ci/builds Builds 657 and 656 each only took ~452 to fail, but build 655 took >10 minutes to fail. Looking at the shards, 2 failed quickly trying to checkout the now-nuked sha. It looks like a fail-fast matrix could fail a job like this faster letting our next job start more quickly.