hi Hatem, There are various causes of flakiness in builds (for example, conda package manager timeouts) -- we have done our best so far to manage the flakiness and will continue to need to manage it. In such cases, you can git commit --amend && git push -f to restart builds without having to make code changes
Generally speaking if you are working on a patch you should only only be concerned with builds that are "downstream" in the dependency graph. Travis CI skips builds that aren't affected, so this should be already what you are seeing. If you see a persistent failure in CI (either in master or in PRs) please open a JIRA issue. We started having failures in the last 72 hours because of https://github.com/apache/arrow/commit/3e97ca1c207cacfb5340940bc86f95107849cbcc, for example, which was caused by an upstream package (pandas) having a release. - Wes On Mon, Jan 14, 2019 at 10:27 AM Hatem Helal <hatem.he...@mathworks.co.uk> wrote: > > Hi everyone, > > I’ve had some trouble getting a clean CI build for a recent pull request and > wanted to understand how this project manages CI failures. I don’t have much > experience with Travis CI to draw on but my own approach so far has been to > try and correlate failures on my own PR against other branches. This helps > me understand whether my changes have broken something. Would it also be > helpful to open JIRA tickets for failures seen on master? It’s also not > obvious to me whether some of the failures might be sporadic in nature and > how those would be best tracked by this project. For example, the failure > might go away if we were to restart the build but that would just temporarily > hide the problem. It would be great to share any best practices or tips that > you have for working with the CI system for this project. > > Thanks, > > Hatem
