[
https://issues.apache.org/jira/browse/SOLR-6002?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13977662#comment-13977662
]
Mark Miller commented on SOLR-6002:
-----------------------------------
Thanks for the review!
bq. 1) I don't understand the need for the CLOSE_LOCK, can you explain?
Just so that only one thread will actually do the close logic. You can close an
IndexWriter with either rollback or close and one of those might internally
call the other - so the CLOSE_LOCK just makes sure the SolrIndexWriter close
logic happens once. We don't use the this lock just so that we don't end up
conflicting with other uses.
bq. 2) In DirectUpdateHandler2 why the change from writer.shutdown() to
writer.close()? All the other changes go in the other direction, e.g. close ->
shutdown.
Because shutdown will wait for merges and commit, etc and DirectUpdateHandler2
actually does that right above - optionally - it was actually a small bug that
someone changed it to shutdown (that's just on branch 5x though). We didn't
want to commit in all cases like shutdown was doing.
I do see a problem though - we need to call waitForMerges before we close.
> Fix a couple of ugly issues around SolrIndexWriter close and rollback as well
> as how SolrIndexWriter manages it's ref counted directory instance.
> -------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: SOLR-6002
> URL: https://issues.apache.org/jira/browse/SOLR-6002
> Project: Solr
> Issue Type: Bug
> Reporter: Mark Miller
> Assignee: Mark Miller
> Priority: Critical
> Fix For: 4.9, 5.0
>
> Attachments: SOLR-6002.patch, SOLR-6002.patch, SOLR-6002.patch
>
>
> We tried jumping all kinds of hoops (unsuccessfully anyway) to *really*
> *really* close an IndexWriter in the past, but Lucene wouldn't let you. Now
> that is fixed in Lucene and we should update our close/rollback code.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
