[
https://issues.apache.org/jira/browse/SOLR-2861?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13139785#comment-13139785
]
Yonik Seeley commented on SOLR-2861:
------------------------------------
Update: I think I may have found one logic bug in solr - a hard commit reopen
can go back in time since it doesn't use the writer to repoen. For example
Thread1: commit() Thread2: reopen(writer) Thread1: reopen(without_writer).
That last reopen will reopen at the point of the commit, going back in time if
updates happened in between.
Now the bad news: I changed the test to only use soft commits, and still hit
problems. I also changed some of the prints to immediately proceed and follow
the lucene API calls to try and further nail this down. The log file was too
large to attach, but here is the analysis:
{code}
13628 TEST:WRITER21: adding id 4 val= 21
13779 VERBOSE:WRITER21: updateDocument id:4
Document<stored,indexed,indexOptions=DOCS_ONLY<id:4>
stored,indexed,tokenized,omitNorms,indexOptions=DOCS_ONLY<val_l:21>>
org.apache.solr.update.SolrIndexWriter@4ba76eff
13969 VERBOSE:WRITER21: updateDocument id:4 DONE
69276 VERBOSE:WRITER3: start reopen from Searcher@17db8f8e
main{DirectoryReader(segments_4:1320005350742:nrt _4h(4.0):C18/17 _4i(4.0):C3/2
_4j(4.0):C1 _4k(4.0):C3/1 _4l(4.0):C6)} writer=
org.apache.solr.update.SolrIndexWriter@4ba76eff
69293 VERBOSE:WRITER3: reopen result
DirectoryReader(segments_4:1320005350743:nrt _4h(4.0):C18/17 _4i(4.0):C3/2
_4j(4.0):C1 _4k(4.0):C3/1 _4l(4.0):C6/1)
69349 TEST:WRITER2: deleting id 4 val= 22
69365 VERBOSE:WRITER2: deleteDocuments id:4
org.apache.solr.update.SolrIndexWriter@4ba76eff
69366 VERBOSE:WRITER9: start reopen from Searcher@256d6cf
main{DirectoryReader(segments_4:1320005350743:nrt _4h(4.0):C18/17 _4i(4.0):C3/2
_4j(4.0):C1 _4k(4.0):C3/1 _4l(4.0):C6/1)} writer=
org.apache.solr.update.SolrIndexWriter@4ba76eff
69367 VERBOSE:WRITER9: reopen result (null)
69394 VERBOSE:WRITER2: deleteDocuments id:4 DONE
69424 VERBOSE:WRITER7: start reopen from Searcher@61c3e3fb
main{DirectoryReader(segments_4:1320005350743:nrt _4h(4.0):C18/17 _4i(4.0):C3/2
_4j(4.0):C1 _4k(4.0):C3/1 _4l(4.0):C6/1)} writer=
org.apache.solr.update.SolrIndexWriter@4ba76eff
69426 VERBOSE:WRITER7: reopen result (null)
T
69447 VERBOSE:READER14: RealTimeGet using searcher Searcher@742136c6
main{DirectoryReader(segments_4:1320005350743:nrt _4h(4.0):C18/17 _4i(4.0):C3/2
_4j(4.0):C1 _4k(4.0):C3/1 _4l(4.0):C6/1)}
TEST:READER14: ERROR, id= 4 foundVal= 21 model val= -20 realTime= true
{code}
So as you can see, this looks much like anaylsis5 - a delete happened during
the first reopen (hence may or may not be reflected in that reopen), and then
there was another reopen which resulted in "null" (no changes). But a lookup
using the resulting reader finds the document which should be deleted.
This is still looking like we have a lucene bug.
> transaction log / realtime-get failures
> ----------------------------------------
>
> Key: SOLR-2861
> URL: https://issues.apache.org/jira/browse/SOLR-2861
> Project: Solr
> Issue Type: Bug
> Reporter: Yonik Seeley
> Attachments: TEST-org.apache.solr.search.TestRealTimeGet.xml,
> fail4.xml, fail5.xml, jenkins_output.txt, trunk-0.txt
>
>
> realtime-get or the transaction log fails sometimes
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
