[
https://issues.apache.org/jira/browse/SOLR-12739?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16642864#comment-16642864
]
Shalin Shekhar Mangar commented on SOLR-12739:
----------------------------------------------
bq. Explaining users how does maxShardsPerNode interplay with policeis would be
very difficult no?
I don't think so. When both are specified, the stricter one applies. I've added
this sentence to the collections api documentation where maxShardsPerNode is
mentioned
bq. If there was no maxShardsPerNode going forward there would only be one way
to set rules up to distirbute replicas - policies
Yes, that is true but specifying maxShardsPerNode as a parameter is far more
simpler than writing a policy rule so I am inclined to continue to support it.
However, it'd be nice to change the internal implementation to eventually use a
policy rule and stop persisting maxShardsPerNode in the cluster state.
bq. We should also ship with some default policies - "Don't allow more than one
replica of a shard on the same JVM" , "Distribute cores across the cluster
evenly" , "Distribute replicas per collection across the nodes"
+1, I'll create an issue. It is anyway needed because the legacy assignment
took care of spreading replicas of same shard on different nodes but
autoscaling by default does not.
bq. Historically, without autoscaling, that parameter defaults to 1. I think
that's also what it should do with autoscaling.
I agree. In this patch, I've kept the default as 1. If a user wants the old
autoscaling specific default, they can pass {{maxShardsPerNode=-1}}.
The latest patch passes all tests and precommit. I'll commit this now.
> Make autoscaling policy based replica placement the default strategy for
> placing replicas
> -----------------------------------------------------------------------------------------
>
> Key: SOLR-12739
> URL: https://issues.apache.org/jira/browse/SOLR-12739
> Project: Solr
> Issue Type: Improvement
> Security Level: Public(Default Security Level. Issues are Public)
> Components: AutoScaling, SolrCloud
> Reporter: Shalin Shekhar Mangar
> Priority: Major
> Fix For: 7.6, master (8.0)
>
> Attachments: SOLR-12739.patch, SOLR-12739.patch, SOLR-12739.patch,
> SOLR-12739.patch, SOLR-12739.patch
>
>
> Today the default placement strategy is the same one used since Solr 4.x
> which is to select nodes on a round robin fashion. I propose to make the
> autoscaling policy based replica placement as the default policy for placing
> replicas.
> This is related to SOLR-12648 where even though we have default cluster
> preferences, we don't use them unless a policy is also configured.
--
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
