[jira] [Updated] (SOLR-6554) Speed up overseer operations for collections with stateFormat > 1

["Shalin Shekhar Mangar (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/SOLR-6554?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Shalin Shekhar Mangar updated SOLR-6554:
----------------------------------------
    Attachment: SOLR-6554.patch

Here's a better patch which takes care of some of the nocommits:
# The live nodes information is read from the ZkStateReader's cluster state because the one in ZkStateWriter's cluster state might be stale
# We force refresh the clusterstate once at the beginning of the loop and then only if there's an error in the main loop.
# Added a simple TestClusterStateMutator, I'll add more.

I still have a few ideas I'd like to try.

> Speed up overseer operations for collections with stateFormat > 1
> -----------------------------------------------------------------
>
>                 Key: SOLR-6554
>                 URL: https://issues.apache.org/jira/browse/SOLR-6554
>             Project: Solr
>          Issue Type: Improvement
>          Components: SolrCloud
>    Affects Versions: 5.0, Trunk
>            Reporter: Shalin Shekhar Mangar
>         Attachments: SOLR-6554.patch, SOLR-6554.patch
>
>
> Right now (after SOLR-5473 was committed), a node watches a collection only if stateFormat=1 or if that node hosts at least one core belonging to that collection.
> This means that a node which is the overseer operates on all collections but watches only a few. So any read goes directly to zookeeper which slows down overseer operations.
> Let's have the overseer node watch all collections always and never remove those watches (except when the collection itself is deleted).



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org