[ 
https://issues.apache.org/jira/browse/SOLR-11478?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16205397#comment-16205397
 ] 

Cao Manh Dat commented on SOLR-11478:
-------------------------------------

For shutdown case, this change makes sense and can easily get done. But for 
startup cases, this is not trivial and bring little benefit
- When the node startup, it will register all its replicas as DOWN, therefore 
we do not forward any requests to replicas which have state = DOWN, although 
/live_nodes entry is already created.
- Replicas will want to receive updates requests when its state = RECOVERING ( 
but refuse query requests ) on startup, therefore live_nodes entry must already 
be created.

So advice for anyone is looking at both /live_nodes and state of replica before 
sending commands to them.

> Solr should remove it's entry from live_nodes in zk immediately on shutdown 
> and add after solr has loaded it's cores and is ready to serve requests.
> ----------------------------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: SOLR-11478
>                 URL: https://issues.apache.org/jira/browse/SOLR-11478
>             Project: Solr
>          Issue Type: Improvement
>      Security Level: Public(Default Security Level. Issues are Public) 
>          Components: SolrCloud
>            Reporter: Binoy Dalal
>            Priority: Minor
>
> Solr currently, upon receiving the stop command, removes its entry from the 
> zk /live_nodes znode after it has finished processing all inflight requests, 
> just before finally shutting down.
> In this case, any applications that depend on a solr node's live_node entry 
> to decide whether or not to query it fail once the stop command has been 
> executed but solr has not yet fully shut down.
> Something similar occurs during startup of a solr node. The solr node seems 
> to add it's entry to the /live_nodes in zk once it is up but before it has 
> started accepting requests and once again, this causes dependent applications 
> to fail in a similar fashion.
> Hence, removal of the node entry and addition of the same to the zk 
> live_nodes immediately upon shutting down and at the very end upon coming up 
> respectively will greatly benefit applications that depend the live_nodes 
> znode.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to