----- Original Message ----- > Is there anything else we can do to prevent this from happening again?
It should be possible to track the peak number of live DOM windows in a test. The current number of live windows is printed out in the log as the number right after --DOMWINDOW == or ++DOMWINDOW == . >From there, the question is what to do with that data. One possibility would >be to track it on tree-management. Somebody would have to keep an eye on it. >New tests that cause this number to go up by a lot could be backed out. I >feel like in the end, the real problem keeping the tree green was a handful of >individual tests that were creating hundreds of windows, so the addition of >tests like that should be fairly easy to do. The problem getting slowly worse >until suddenly it falls over is a concern, but maybe because we're tracking >the maximum it is less of an issue? Another way to expose this information would be to probabilistically make the test orange based on how close we are to some per-test peak window limit. The idea here as we slowly creep closer to the limit, the test would slowly become more orange. At a certain point, sheriffs would notice, and get somebody to look into what is happening, before the tree falls over entirely. You'd want to track this orange on a per-test-suite basis (I guess by including the name of the suite in the orange), and tuning the probability function would be tricky. Plus the sheriff's may not be a fan of graduated-sheriff-annoyance as a strategy, so maybe the more conventional tree-management mail would be better. Andrew _______________________________________________ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
