Hi, Petr, Thanks for the test data. Looks like the problem is caused by a set of shared buffers that need to be made private (not shared). The change has been checked in as SVN Revision 586. Please take a look when you get the chance.
John On 10/2/12 2:11 AM, Thorgrin wrote: > Hi John, > > thanks for the bugfix. I just tested the new SVN revision 584 and the > segfault is gone. However, there seems to be some kind of problem with > aggregation on long columns when multiple partitions are involved (I'm > able to reproduce the error using multiple partitions, maybe it also > occurs using one partition). The aggregation result gives wrong > result. I included test data so that you can see for yourself. > > The commands to use: > thula -d ./test -s "e0id4,sum(e0id1),sum(e0id2)" -w "1=1" > thula -d ./test -s "e0id4,sum(e0id1),sum(e0id2),sum(e0id152)" -w "1=1" > > You can see that the first two summary column give different results > when aggregation on long columns is added. > > Best regards, > Petr > > > On 1 October 2012 21:25, K. John Wu <[email protected]> wrote: >> Hi, Gaurav >> >> Sorry for missing that - I see that you have included the content of >> the file in the original message. Please give SVN Revision 584 a try >> when you get the chance. Let me know if you spot anything else. >> >> Thanks. >> >> John >> >> >> On 10/1/12 10:58 AM, Gaurav Agarwal wrote: >>> Hi John, >>> >>> The contents of the CSV file is mentioned in the original mail - it is >>> a toy file with 3 rows and 2 columns: >>> >>> --Test Data-- >>> $ cat ~/Desktop/tmp.txt >>> a,1 >>> b,2 >>> c,3 >>> >>> cheers, >>> gaurav >>> >>> On Mon, Oct 1, 2012 at 9:12 PM, K. John Wu <[email protected] >>> <mailto:[email protected]>> wrote: >>> >>> Hi, Gaurav, >>> >>> Would you be willing to share your CSV file? Thanks. >>> >>> John >>> >>> >>> On 9/30/12 9:38 PM, Gaurav Agarwal wrote: >>> > Hi John, >>> > >>> > There is a problem with the current SVN trunk - if I try to do a sum >>> > of long values broken down by a column, there is a segmentation >>> fault. >>> > If I change the column type to double from long, the query works >>> fine. >>> > Here are the steps to reproduce it: >>> > >>> > --Test Data-- >>> > $ cat ~/Desktop/tmp.txt >>> > a,1 >>> > b,2 >>> > c,3 >>> > >>> > --Import-- >>> > $ardea -d /tmp/tst -t ~/Desktop/tmp.txt -m "c1:k,c2:l" >>> > ardea: verbose level 0 >>> > Will attempt to parse 1 CSV file >>> > /Users/gaurav/Desktop/tmp.txt >>> > with the following column names and types >>> > c1:k,c2:l >>> > >>> > >>> > ardea to read CSV file /Users/gaurav/Desktop/tmp.txt ... >>> > ardea read 3 rows from /Users/gaurav/Desktop/tmp.txt >>> > >>> > --Query-- >>> > $thula -d /tmp/tst -s "c1,sum(c2)" -w "c1!='foo'" -v 2 >>> > >>> > FastBit ibis1.3.2.6 >>> > Log messages started on Mon Oct 1 10:08:05 2012 >>> > util::gatherParts -- examining /tmp/tst >>> > Constructed a part named tst >>> > activeDir = "/tmp/tst" >>> > 3 rows and 2 columns >>> > mensa::addPartition(/tmp/tst) found 1 new data partition >>> > mensa::addPartition(/tmp/tst) increases the number partitions from 0 >>> > to 1, the number of rows from 0 to 3, and the number of columns >>> from 0 >>> > to 2 >>> > Table (on disk) T-tst (/tmp/tst) consists of 1 partition with 2 >>> > columns and 3 rows >>> > c1CATEGORY >>> > c2LONG >>> > thula >>> > Select c1,sum(c2) >>> > From T-tst >>> > Where -- >>> > c1!='foo' >>> > countQuery::setWhereClause -- add a new where clause "( ! >>> 0x10db022d0)" >>> > countQuery assigned data partition tst >>> > countQuery::estimate -- number of hits is 3 >>> > countQuery::estimate -- duration: 0.00044 sec(CPU), 0.000439882 >>> > sec(elapsed) >>> > doQuery(c1!='foo') -- the estimated number of hits on T-tst is 3 >>> > countQuery::setWhereClause accepted new query conditions "( ! >>> > 0x10db01640)" >>> > Constructed in-memory data partition guhXd0 -- filter::sift2(SELECT >>> > c1,sum(c2) FROM 1 data partition WHERE ( ! 0x ...) -- with 2 columns >>> > filter::sift2(SELECT c1,sum(c2) FROM 1 data partition WHERE ( ! 0x >>> > ...) -- processing data partition tst >>> > countQuery assigned data partition tst >>> > countQuery::evaluate -- Select count(*) From tst Where ( ! >>> > 0x10db01640) --> 3 >>> > countQuery::evaluate -- duration: 7.8e-05 sec(CPU), 7.79629e-05 >>> > sec(elapsed) >>> > Mon Oct 1 10:08:05 2012 >>> > Warning -- column[guhXd0.__1](LONG)::selectDoubles -- incompatible >>> > data type >>> > Segmentation fault: 11 >>> > >>> > cheers, >>> > gaurav >>> > >>> > >>> > _______________________________________________ >>> > FastBit-users mailing list >>> > [email protected] <mailto:[email protected]> >>> > https://hpcrdm.lbl.gov/cgi-bin/mailman/listinfo/fastbit-users >>> > >>> >>> >> _______________________________________________ >> FastBit-users mailing list >> [email protected] >> https://hpcrdm.lbl.gov/cgi-bin/mailman/listinfo/fastbit-users >> >> >> _______________________________________________ >> FastBit-users mailing list >> [email protected] >> https://hpcrdm.lbl.gov/cgi-bin/mailman/listinfo/fastbit-users _______________________________________________ FastBit-users mailing list [email protected] https://hpcrdm.lbl.gov/cgi-bin/mailman/listinfo/fastbit-users
