Hi, Made some updates, as I was able to grab an Ubuntu box. The build seems to fail, consistently, on ubuntu. Regarding the NPE, the SearchManager (tsp on the test) is *not* null, but the attachment is. Oddly this only happens on ubuntu.
I'll try to debug some more, but I've limited access to this box, so extra eyes are more than welcome. br, juan pablo El mar., 7 may. 2019 0:24, Juan Pablo Santos Rodríguez < juanpablo.san...@gmail.com> escribió: > My daily update on this: > > definitely, seems something is weird on the ubuntu nodes; I've been trying > several combinations of JDK 8, maven installations, increasing the -Xmx > through MAVEN_OPTS, and all failed, on ubuntu nodes. I've modified the > build to run on windows nodes, and suddenly there aren't any problems. I've > also tried on my local pc (windows based) with several JDK 8 and memory > settings and haven't been able to reproduce the test failure. > > I'll try tomorrow on an Ubuntu box at work to see if I can reproduce the > test failure. Meanwhile, has anybody been able to reproduce it? > What do we do with the build process? Comment out the test, move it to > Windows nodes,..? > > > best regards, > juan pablo > > On Sun, May 5, 2019 at 11:44 PM Juan Pablo Santos Rodríguez < > juanpablo.san...@gmail.com> wrote: > >> Hi, >> >> hmm, didn't realize that Tika and JSPWiki pull different versions (7.5.0 >> and 8.0.0). However, JSPWiki "forces" it's Lucene version to >> Tika via dependencyManagement section in pom, so when pulling the >> dependencies only Lucene 8 appears in classpath for both >> JSPWiki and Tika. We could have binary incompatibilities down the line, >> but I think we are not using Tika enough for them to appear. >> Also, there are occasions of tests failing where Tika isn't even loaded >> in memory (https://builds.apache.org/job/jspwiki-test/18/console). >> Tests which have been running for ages, and with NPEs on objects that are >> obviously not null at that point.. Tika related code runs ok, >> the error comes on an NPE afterwards. >> >> Weirder, the offending test fails with: >> >> [ERROR] testGetAttachmentContent Time elapsed: 3.892 s <<< ERROR! >> java.lang.NullPointerException >> at >> org.apache.wiki.search.tika.TikaSearchProviderTest.testGetAttachmentContent(TikaSearchProviderTest.java:58) >> >> being the offending line: >> String pdfIndexed = tsp.getAttachmentContent( attPdf ); >> >> and tsp being loaded as: >> TikaSearchProvider tsp = ( TikaSearchProvider >> )engine.getSearchManager().getSearchEngine(); >> >> (so, only tsp could yield the NPE) >> >> I modified the Jenkinsfile to perform a sed on that test so the above >> line gets replaced by: >> TikaSearchProvider tsp = new TikaSearchProvider();tsp.initialize( engine, >> props ); >> >> tsp *cannot* be null beyond that point.. But it is: >> https://builds.apache.org/job/jspwiki-test/19/ shows the same NPE on the >> same >> line yielding it as before. >> >> Perhaps is that the build is running with a very adjusted memory setting >> (i.e.: the default maven setting is not enough for the jspwiki >> build?). Will keep digging tomorrow.. >> >> >> best regards, >> juan pablo >> >> On Sun, May 5, 2019 at 10:50 PM Murray Altheim <murra...@altheim.com> >> wrote: >> >>> Hi Juan Pablo, >>> >>> My suspicion is still related to there being two Lucene engines in the >>> mix. >>> >>> If JSPWiki and Tika use *different* versions of Lucene there will be two >>> Lucenes the classpath, and I've personally had all sorts of issues with >>> Lucene instantiating when two components both use it (I seem to remember >>> the "Version" can't be resolved, or something like that). Put it this >>> way: >>> I've noted that two Lucene engines in the same project might not like >>> each >>> other too much. I think I might have even had problems if they use the >>> same version number there can be contention. So when I heard something in >>> Tika related to Lucene was popping an NPE it set off my Lucene Alarm... >>> >>> But I've got some time this week so I'll be trying out the latest >>> release, >>> hopefully in the next few days. >>> >>> Cheers, >>> >>> Murray >>> >>> On 6/05/19 8:42 AM, Juan Pablo Santos Rodríguez wrote: >>> > Hi Murray, >>> > >>> > thanks for looking into it :-) >>> > >>> > My findings so far: Lucene index related test are already in the main >>> > build, so that part should be covered, and regarding the Tika-related >>> ones, >>> > they don't interact with Lucene, they only parse a couple of files, and >>> > check the extracted information (which is afterwards sent to Lucene, >>> but >>> > that's outside the scope of the test). I was suspecting of something >>> weird >>> > on the node in which the build kept running and made a new Jenkins job >>> to >>> > do some tests (and avoid spamming dev@j.a.o) at >>> > https://builds.apache.org/job/jspwiki-test/ >>> > [...] >>> >>> ........................................................................... >>> Murray Altheim <murray18 at altheim dot com> = = >>> === >>> http://www.altheim.com/murray/ === >>> === >>> = = >>> === >>> In the evening >>> The rice leaves in the garden >>> Rustle in the autumn wind >>> That blows through my reed hut. >>> -- Minamoto no Tsunenobu >>> >>>
