Another note: these timings are from server-class workstations. On a laptop the overall test length (without the change) can run 18+ hours.
Anthony > On Sep 4, 2015, at 12:21 PM, Kirk Lund <[email protected]> wrote: > > +1 > > I like the idea of switching to fork=1 for a few months to focus on > stabilizing any dunit tests that fail without potential test pollution > causes. These failures are mostly like ones that involve race conditions. > Once we fix these, then we could change back to fork=30. > > > On Fri, Sep 4, 2015 at 11:16 AM, Mark Bretl <[email protected]> wrote: > >> I see that Kirk made an update to the issue and wanted to follow up on the >> Dev list for discussion. >> >> I also ran a build on the open side with the dunit fork = 1. The total time >> of the build was: 9 hrs 58 mins 33.662 secs. The last Geode build took: 6 >> hr 21 min <https://builds.apache.org/job/Geode-nightly/buildTimeTrend>. >> It is an increase of about 3.5 hours, which matches the increase in time >> that Kirk had. >> >> The question becomes: Do we want to make the change so we can increase the >> quality of tests by isolating each one at the cost of increasing the total >> build time? >> >> My thoughts would be to make the change to weed out the 'bad' tests and >> increase the overall quality of the tests, so when a test fails there is no >> questioning the result. Once we have them passing more consistently, then >> we can increase the fork count again. >> >> Thoughts? >> >> -- >> Mark Bretl >> Software Build Engineer >> Pivotal >> 503-533-3869 >> www.pivotal.io >>
