Hi Alexander, I just noticed today something interesting while looking at the console logs that could explain some of the failures. It seems that sometimes, 2 Hudson executors can share the same display which seems like a bug (in the Xvnc Hudson plugin?). For example,
https://hudson.eclipse.org/linuxtools/job/linuxtools-gerrit/9140/consoleFull and https://hudson.eclipse.org/linuxtools/job/linuxtools-gerrit/9141/consoleFull They were running at the same time. In build #9140, the display was :76 ... [workspace_3] $ Xvnc :76 -geometry 1024x768 -depth 24 -ac ... In the build #9141, the same command is executed but with error: [workspace] $ Xvnc :76 -geometry 1024x768 -depth 24 -ac Fatal server error: Server is already active for display 76 If this server is no longer running, remove /tmp/.X76-lock and start again. At the end of both console log, I can see that they ended up both using the same display: Xlib: extension "RANDR" missing on display ":76.0". Then they both failed almost at the exact the same time (16:50:37 and 16:50:32). I'm not sure whether this is a configuration issue but I think this is something worth investigating. Marc-Andre ________________________________________ From: linuxtools-dev-boun...@eclipse.org [linuxtools-dev-boun...@eclipse.org] on behalf of Aleksandar Kurtakov [akurt...@redhat.com] Sent: Saturday, 27 September 2014 3:48 AM To: Linux Tools developer discussions Subject: [linuxtools-dev] Announcement: SWTBot tests I've committed a patch that ignores some of the failing tests. I beg every contributor - PLEASE DO NOT ADD MORE SWTBOT tests. I'm really tired of the instability of these tests. There are many problems with these tests like: * They are fragile - a widget being added/removed in other components breaks them * They are not real unit tests - failures are usually not in the component that is supposed to be under test but elsewhere. * They do not help shaping good API as people don't use the API that are exposed but do "clicks" * So much noise generated by swtbot tests in hudson that make the mailing list unusable * No one cares about test failures anymore and ignores them * ............... I do not mind having SWTBot tests in the codebase but they should not be enabled in the default build and hudson. Every test that runs by default must work 100% when doing "mvn clean verify". >From now on, whenever I spot a test case failing more than twice in gerrit >when testing unrelated patch I'll add ignore to it as it is high time to get >reliable builds. I'm really getting tired of these issues. Tests that fail randomly are useless and even harmless. Period. Alexander Kurtakov Red Hat Eclipse team _______________________________________________ linuxtools-dev mailing list linuxtools-dev@eclipse.org To change your delivery options, retrieve your password, or unsubscribe from this list, visit https://dev.eclipse.org/mailman/listinfo/linuxtools-dev _______________________________________________ linuxtools-dev mailing list linuxtools-dev@eclipse.org To change your delivery options, retrieve your password, or unsubscribe from this list, visit https://dev.eclipse.org/mailman/listinfo/linuxtools-dev
