Bugs item #1975133, was opened at 2008-05-27 18:12 Message generated for change (Comment added) made by jflokstra You can respond by visiting: https://sourceforge.net/tracker/?func=detail&atid=482468&aid=1975133&group_id=56967
Please note that this message will contain a full copy of the comment thread, including the initial issue submission, for this request, not just the latest update. Category: PFtijah Group: Pathfinder 0.24 Status: Open Resolution: None Priority: 6 Private: No Submitted By: Stefan Manegold (stmane) Assigned to: Jan Flokstra (jflokstra) Summary: PF/tijah: most tests fail with property error Initial Comment: most PF/tijah tests fail (with both MPS & ALG back-end) due to property errors like !WARNING: BATpropcheck: BAT tj_testcoll1_pfpre(1264)[oid,oid] with 48 tuples was incorrectly marked keyed! e.g.: http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora8/modules_pftijah/colltest2.err.00.html http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora8/modules_pftijah/qmil0.out.00.html http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTestsG103/GNU.64.64.d-Fedora8/modules_pftijah/colltest2.err.00.html http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTestsG103/GNU.64.64.d-Fedora8/modules_pftijah/qmil0.out.00.html ---------------------------------------------------------------------- >Comment By: Jan Flokstra (jflokstra) Date: 2008-05-29 14:17 Message: Logged In: YES user_id=1054297 Originator: NO I checked the lifecycle of the BAT: 1. Creation of the bat: $PF/modules/pftijah/pftijah.mx:613 new(oid,oid).persists(true).bbpname("tj_" + ftiName + "_pfpre"); 2. Use of mmap: $PF/modules/pftijah/pftijah.mx:1139 collBat.find("_pfpre").access(BAT_APPEND).mmap(1); An then precommit(), location is not important. When I remove the line "collBat.find("_pfpre").access(BAT_APPEND).mmap(1);" which looks up the bat which causes the problems and does the access() and mmap() call everything works fine. Doing only the .access() or the .mmap() also works fine. CONCLUSION FOR NIELS: the construction "BAT.access(BAT_APPEND).mmap(1);" causes the problem here. ---------------------------------------------------------------------- Comment By: Niels Nes (nielsnes) Date: 2008-05-28 18:03 Message: Logged In: YES user_id=43556 Originator: NO Jan could you try to figure out the order of bat access, persistency and mmap allocation changes, ie calls to access(BAT_READ/WRITE), commit and mmap? It seems the order in pftijah is different then the currently tested once. This will help us solve this data corruption problem. ---------------------------------------------------------------------- Comment By: Jan Flokstra (jflokstra) Date: 2008-05-28 16:13 Message: Logged In: YES user_id=1054297 Originator: NO This is a problem with the BAT containing the translation from term index to pathfinder index. The handling of this bat has not changed for a year. The problem is the bat at the end of the indexing transaction contains the correct tuples but after commit and the start of the new transaction it is completely filled with 0's. The last time I had a similar problem (2 weeks ago) this was due to the new mmap() handling but we do not use this here. I will try to pinpoint the exact time of the transformation tomorrow. ---------------------------------------------------------------------- You can respond by visiting: https://sourceforge.net/tracker/?func=detail&atid=482468&aid=1975133&group_id=56967 ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Monetdb-bugs mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/monetdb-bugs
