Sadly, yes, this is SimpleText's "speedup", heh. However, the OOM actually occurs when the RIW used by TestNRQ64 calls optimize on .close(). But, this optimize is unused since the test has already pulled its reader. I think a simple workaround would be to close the RIW but never let it call optimize? Eg we could just reach in and close the underlying writer (writer.w.close)... or add a .closeNoOptimize to RIW?
Mike On Sat, Nov 6, 2010 at 6:07 PM, Uwe Schindler <[email protected]> wrote: > Hehe, yes! > > This test creates lots of terms: > As far as I remember, it indexes 10,000 documents in 9 different number > styles and precisionSteps (2, 4, 6, 8): > > But when RANDOM_MULTIPLIER > 1 then much more: > private static final int noDocs = 10000 * RANDOM_MULTIPLIER; > > As the lowerPrec terms are duplicates, the approx. number of terms (I assume > that for next shift value, the number of terms reduces by 2 because of > overlaps): > precStep=8: 10 000 + 5 000 + 2 500 + 1 250 + 625 + 312 + 161 + 80 = 19 928 > terms > precStep=6: 10 000 + 5 000 + 2 500 + 1 250 + 625 + 312 + 161 + 80 + 40 + 20 + > 10 = 19 998 terms > precStep=4: some more (16 summands) :-], approx. 20 500 > precStep=2: again more (32 summands), approx. 20800 > > This makes approx 9*20,000*RANDOM_MULTIPLIER (its 3 on Hudson) terms in the > index -- ahm TreeSet incl TermInfo *g*. For TestNumericRangeQuery32 its > similar, but fewer precsteps. > > We have several options: Run all tests with default Heap size only for > SimpleText and check that all tests pass, if not raise -Xmx until it passes > (currently tests use 512 M). > > Alternatively reduce the index size for some tests if SimpleText codec is > used. TestNumeric should be easy possible, as the test uses a some > preconfigured constants for building index and run tests. > > ----- > Uwe Schindler > H.-H.-Meier-Allee 63, D-28213 Bremen > http://www.thetaphi.de > eMail: [email protected] > > >> -----Original Message----- >> From: Robert Muir [mailto:[email protected]] >> Sent: Saturday, November 06, 2010 9:03 PM >> To: [email protected] >> Subject: Re: Lucene-Solr-tests-only-trunk - Build # 1071 - Still Failing >> >> On Sat, Nov 6, 2010 at 3:58 PM, Apache Hudson Server >> <[email protected]> wrote: >> > Error Message: >> > Java heap space >> > >> ... >> > Error Message: >> > MockDirectoryWrapper: cannot close: there are still open files: >> > {_2h.pst=1} >> >> seems like its likely caused by the simpletext optimization? >> >> --------------------------------------------------------------------- >> To unsubscribe, e-mail: [email protected] For additional >> commands, e-mail: [email protected] > > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > > --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
