[ 
https://issues.apache.org/jira/browse/SOLR-12067?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16393730#comment-16393730
 ] 

Shalin Shekhar Mangar commented on SOLR-12067:
----------------------------------------------

Thanks Varun. I agree 30 seconds is less. Actually, I found that with HDFS the 
timeout was autoReplicaFailoverBadNodeExpiration (default 60s) + 
autoReplicaFailoverWaitAfterExpiration (default 30s). We deprecated 
autoReplicaFailoverBadNodeExpiration value but did not add it to the default 
autoReplicaFailoverWaitAfterExpiration. So the timeout should be 90 seconds at 
least. I think we should be conservative here and set this to a higher value, 
say 120s.

> AutoAddReplicas default 30 second wait time is too low
> ------------------------------------------------------
>
>                 Key: SOLR-12067
>                 URL: https://issues.apache.org/jira/browse/SOLR-12067
>             Project: Solr
>          Issue Type: Bug
>      Security Level: Public(Default Security Level. Issues are Public) 
>            Reporter: Varun Thacker
>            Assignee: Shalin Shekhar Mangar
>            Priority: Major
>
> If I create a collection with autoAddReplicas=true in Solr 7.x a 
> AutoAddReplicasPlanAction get's created with a waitFor=30 seconds
> The default should be increased as a JVM which is down for more than 
> 30seconds can cause the framework to add a new replica on another node
> With HDFS this was a cheap operation as it only involved create a core and 
> pointing it to the same index directory. 
> But for non shared file systems, this is a very expensive operation and can 
> potentially move large indexes around so maybe we should have a higher default



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

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

Reply via email to