[jira] Commented: (LUCENE-2907) termsenum bug when running with multithreaded search
[ https://issues.apache.org/jira/browse/LUCENE-2907?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=12991133#comment-12991133 ] Robert Muir commented on LUCENE-2907: - bq. Have you found out what happens or where a thread-safety issue could be? Yes, i found the bug... unfortunately it is actually my automaton problem :( I will create a nice patch today. bq. The information on this issue is too small, there seems to be lots of IRC/GTalk communication in parallel. what do you mean? mike was working a long time on the bug, but quickly had to stop working on it, so he emailed me all of his state. I took over from there for a while, and i opened this issue with my debugging... though I didn't have much time to work on it yesterday (only like 1 hour), because I already had plans. I tried to be completely open and dump all of my state/debugging information/brainstorming on this JIRA issue, but it only resulted in me reporting misleading and confusing information... so I think the information on this issue is actually too much? termsenum bug when running with multithreaded search Key: LUCENE-2907 URL: https://issues.apache.org/jira/browse/LUCENE-2907 Project: Lucene - Java Issue Type: Bug Reporter: Robert Muir Attachments: LUCENE-2907_repro.patch, correct_seeks.txt, incorrect_seeks.txt, seeks_diff.txt This one popped in hudson (with a test that runs the same query against fieldcache, and with a filter rewrite, and compares results) However, its actually worse and unrelated to the fieldcache: you can set both to filter rewrite and it will still fail. -- This message is automatically generated by JIRA. - For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] Commented: (LUCENE-2907) termsenum bug when running with multithreaded search
[ https://issues.apache.org/jira/browse/LUCENE-2907?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=12991137#comment-12991137 ] Uwe Schindler commented on LUCENE-2907: --- A bug in automaton that only hapoens in multi-threaded? So its the cache there? termsenum bug when running with multithreaded search Key: LUCENE-2907 URL: https://issues.apache.org/jira/browse/LUCENE-2907 Project: Lucene - Java Issue Type: Bug Reporter: Robert Muir Attachments: LUCENE-2907_repro.patch, correct_seeks.txt, incorrect_seeks.txt, seeks_diff.txt This one popped in hudson (with a test that runs the same query against fieldcache, and with a filter rewrite, and compares results) However, its actually worse and unrelated to the fieldcache: you can set both to filter rewrite and it will still fail. -- This message is automatically generated by JIRA. - For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] Commented: (LUCENE-2907) termsenum bug when running with multithreaded search
[ https://issues.apache.org/jira/browse/LUCENE-2907?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=12991140#comment-12991140 ] Robert Muir commented on LUCENE-2907: - in combination with other things. in my opinion the problem is the cache in getNumberedStates. But the real solution (in my opinion) is to clean up all this crap so the termsenum only takes a completely immutable view of what it needs and for the Query to compile once in its ctor, and remove any stupid caching. So, this is what I am working on now. termsenum bug when running with multithreaded search Key: LUCENE-2907 URL: https://issues.apache.org/jira/browse/LUCENE-2907 Project: Lucene - Java Issue Type: Bug Reporter: Robert Muir Attachments: LUCENE-2907_repro.patch, correct_seeks.txt, incorrect_seeks.txt, seeks_diff.txt This one popped in hudson (with a test that runs the same query against fieldcache, and with a filter rewrite, and compares results) However, its actually worse and unrelated to the fieldcache: you can set both to filter rewrite and it will still fail. -- This message is automatically generated by JIRA. - For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org