[
https://issues.apache.org/jira/browse/LUCENE-9302?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17073151#comment-17073151
]
Ishan Chattopadhyaya commented on LUCENE-9302:
----------------------------------------------
Thanks David, moved to a Lucene issue now. Would like some early feedback on
the patch here. Mainly curious whether there's some reason why the
totalHitCounts is an integer, not a long? Given that the merge() method's
javadocs say the following, I feel the total hits across all shards can
legitimately overflow the integer range (this is happening in our production
cluster).
{code:java}
/** Merges an array of TopGroups, for example obtained
* from the second-pass collector across multiple
* shards. Each TopGroups must have been sorted by the
* same groupSort and docSort, and the top groups passed
* to all second-pass collectors must be the same.
*
* <b>NOTE</b>: We can't always compute an exact totalGroupCount.
* Documents belonging to a group may occur on more than
* one shard and thus the merged totalGroupCount can be
* higher than the actual totalGroupCount. In this case the
* totalGroupCount represents a upper bound. If the documents
* of one group do only reside in one shard then the
* totalGroupCount is exact.
*
* <b>NOTE</b>: the topDocs in each GroupDocs is actually
* an instance of TopDocsAndShards
*/
public static <T> TopGroups<T> merge(TopGroups<T>[] shardGroups, Sort
groupSort, Sort docSort, int docOffset, int docTopN, ScoreMergeMode
scoreMergeMode) {
{code}
> Integer overflow in total count in grouping results
> ---------------------------------------------------
>
> Key: LUCENE-9302
> URL: https://issues.apache.org/jira/browse/LUCENE-9302
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Ian
> Assignee: Ishan Chattopadhyaya
> Priority: Minor
> Attachments: SOLR-13004.patch, SOLR-13004.patch
>
>
> When doing a Grouping search in solr cloud you can get a negative number for
> the total found.
> This is caused by the accumulated total being held in an integer and not a
> long.
>
> example result:
> {{{ "responseHeader": { "status": 0, "QTime": 9231, "params": { "q":
> "decade:200", "indent": "true", "fl": "decade", "wt": "json", "group.field":
> "decade", "group": "true", "_": "1542773674247" } }, "grouped": { "decade": {
> "matches": -629516788, "groups": [ { "groupValue": "200", "doclist": {
> "numFound": -629516788, "start": 0, "maxScore": 1.9315376, "docs": [ {
> "decade": "200" } ] } } ] } } }}}
>
> {{result without grouping:}}
> {{{ "responseHeader": { "status": 0, "QTime": 1063, "params": { "q":
> "decade:200", "indent": "true", "fl": "decade", "wt": "json", "_":
> "1542773791855" } }, "response": { "numFound": 3665450508, "start": 0,
> "maxScore": 1.9315376, "docs": [ { "decade": "200" }, { "decade": "200" }, {
> "decade": "200" }, { "decade": "200" }, { "decade": "200" }, { "decade":
> "200" }, { "decade": "200" }, { "decade": "200" }, { "decade": "200" }, {
> "decade": "200" } ] } }}}
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]