[jira] [Updated] (SOLR-5875) QueryComponent.mergeIds() unmarshals all docs' sort field values once per doc instead of once per shard
[
https://issues.apache.org/jira/browse/SOLR-5875?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Steve Rowe updated SOLR-5875:
-
Attachment: SOLR-5875.patch
Simple patch with fix.
[~alexey] has confirmed that this solved the excessive memory use issue he saw.
Committing shortly.
QueryComponent.mergeIds() unmarshals all docs' sort field values once per doc
instead of once per shard
---
Key: SOLR-5875
URL: https://issues.apache.org/jira/browse/SOLR-5875
Project: Solr
Issue Type: Bug
Components: search
Affects Versions: 4.7
Reporter: Steve Rowe
Assignee: Steve Rowe
Priority: Critical
Fix For: 4.7.1
Attachments: SOLR-5875.patch
SOLR-5354 added unmarshalling of distributed sort field values in
{{QueryComponent.mergeIds()}}, but incorrectly performs this (unmarshalling
all docs' sort field values) for every doc, and stores the result with each
doc. This is unnecessary, inefficient, and extremely wasteful of memory.
In an offline conversation, [~alexey] described the issue to me and located
the likely problem, and [~hossman_luc...@fucit.org] located the problem code
via inspection.
This bug is very likely the problem described on the solr-user mailing list
here: [SolrCloud constantly crashes after upgrading to Solr
4.7|http://mail-archives.apache.org/mod_mbox/lucene-solr-user/201403.mbox/%3c83f549bdf8deecbc7567c324ee0cb...@cluster38.e-active.nl%3e]
--
This message was sent by Atlassian JIRA
(v6.2#6252)
-
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Updated] (SOLR-5875) QueryComponent.mergeIds() unmarshals all docs' sort field values once per doc instead of once per shard
[
https://issues.apache.org/jira/browse/SOLR-5875?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Steve Rowe updated SOLR-5875:
-
Fix Version/s: 5.0
4.8
QueryComponent.mergeIds() unmarshals all docs' sort field values once per doc
instead of once per shard
---
Key: SOLR-5875
URL: https://issues.apache.org/jira/browse/SOLR-5875
Project: Solr
Issue Type: Bug
Components: search
Affects Versions: 4.7
Reporter: Steve Rowe
Assignee: Steve Rowe
Priority: Critical
Fix For: 4.8, 5.0, 4.7.1
Attachments: SOLR-5875.patch
SOLR-5354 added unmarshalling of distributed sort field values in
{{QueryComponent.mergeIds()}}, but incorrectly performs this (unmarshalling
all docs' sort field values) for every doc, and stores the result with each
doc. This is unnecessary, inefficient, and extremely wasteful of memory.
In an offline conversation, [~alexey] described the issue to me and located
the likely problem, and [~hossman_luc...@fucit.org] located the problem code
via inspection.
This bug is very likely the problem described on the solr-user mailing list
here: [SolrCloud constantly crashes after upgrading to Solr
4.7|http://mail-archives.apache.org/mod_mbox/lucene-solr-user/201403.mbox/%3c83f549bdf8deecbc7567c324ee0cb...@cluster38.e-active.nl%3e]
--
This message was sent by Atlassian JIRA
(v6.2#6252)
-
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
