On 20/09/2013 11:02, Rainer Jung wrote: > Having run some sporadically failing tests in a loop under the three > connectors on Solaris 10 Sparc using Java 1.7.0_40. Code is trunk > r1524838 (current as of this morning and codewise identical to 8.0.0 RC3).
Great. This is really helpful. If you could repeat the test with the various fixes in place to see if things have got any better that would be great. > The system is relatively slow but not *that* slow. OTOH it had some NFS > server duties while I was running the tests. > > > Overview > ======== > > > TestWebappClassLoaderExecutorMemoryLeak > --------------------------------------- > > bio: fails 2/10 > nio: hangs 1/10 > apr: fails 3/10 > > The hang for nio in the leak detection test looks suspect (see below), > the other failures might be uncritical timing issues: Looking at the code, timing issues in the tests look very likely. I've committed a fix for the timing issues. Regarding the loop, any ideas on the root cause? > TestWsSubprotocols > ------------------ > > bio: fails 3/25 > nio: fails 1/25 > apr: fails 3/25 > > Testcase: testWsSubprotocols took 5.834 sec > Caused an ERROR > null > java.lang.NullPointerException > at > org.apache.tomcat.websocket.TestWsSubprotocols.testWsSubprotocols(TestWsSubprotocols.java:89) > > NPEs get logged for some of the failures, see below. I think this is another timing error if the client moves faster than the server. I can replicated this failure with a breakpoint on the list that sets the static subprotocols field. I've committed a fix for this as well. > TestWsWebSocketContainer > ------------------------ > > bio: 2/15 > nio: 0/15 > apr: 0/15 > > Crash for APR seems fixed (or very rare) :) > Details for bio failures see below. I'm going to assume that the crash is fixed for now. I can see how the BIO errors occur. If the session is closed before the message is fully written to the buffers, there will be an IOE and the test doesn't handle that. I've fixed this in svn too. > TestCoyoteAdapter > ----------------- > > bio: 0/10 > nio: 0/10 > apr: 0/10 > > So yesterday's single failure here was a rare case. OK. Personally, I'm prepared to accept the odd rare case. Mark --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org For additional commands, e-mail: dev-h...@tomcat.apache.org
