[ 
https://issues.apache.org/jira/browse/SOLR-13226?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16761974#comment-16761974
 ] 

Amrit Sarkar commented on SOLR-13226:
-------------------------------------

My understanding of having non-strict rules is "follow it until there is no 
other way". Unlike here, where those rules are ignored and not taken into 
consideration at all.

bq. Setting the rules as strict may mitigate the problem, but why is Solr doing 
that?
I believe that is how the trigger is designed, of what source code and test I 
read and understood. Do ADDREPLICA operation on any available 
node/shard/collection until a violation occurs. [~shalinmangar] may have 
something to add on this.

> Add note for Node Added Trigger documentation in Autoscaling
> ------------------------------------------------------------
>
>                 Key: SOLR-13226
>                 URL: https://issues.apache.org/jira/browse/SOLR-13226
>             Project: Solr
>          Issue Type: Improvement
>      Security Level: Public(Default Security Level. Issues are Public) 
>          Components: AutoScaling
>            Reporter: Amrit Sarkar
>            Priority: Major
>         Attachments: SOLR-13226.patch, Screen Shot 2019-02-05 at 3.55.31 
> AM.png, Screen Shot 2019-02-05 at 4.02.00 AM.png
>
>
> Node Added Trigger doesn't abide by SOFT rules [strict: false] and results in 
> abnormal cluster operational behavior.
> Let's say; we wish to do the following:
> 1. Not more than 10 cores reside on Single node.
> 2. Wish to distribute the cores, replicas equally to each Node.
> If we go by the following policy:
> not more than one replica for unique shard on a node. not a strict rule.
> {code}
>   {"replica":"<2", "shard": "#EACH", "node": "#ANY", "strict": false},
> {code}
> distribute the replicas equally across the nodes, not a strict rule.
> {code}
>   {"replica": "#EQUAL", "node": "#ANY", "strict": false},
> {code}
> not more than 10 cores allowed on a single node, strict rule.
> {code}
>   {"cores": "<10", "node": "#ANY"}
> {code}
> cluster state ends up like:
> Screenshot -1
> Only the strict rule is followed and multiple replicas are added to single 
> Solr node, as rules are not strict – _{"replica":"<2", "shard": "#EACH", 
> "node": "#ANY", "strict": false}_
> While the following with all strict rule generate normal operational 
> behavior, add a replica to each shard of collection 'wiki' :
> {code}
> [
>   {"replica":"<2", "shard": "#EACH", "node": "#ANY"},
>   {"replica": "#EQUAL", "node": "#ANY"},
>   {"cores": "<10", "node": "#ANY"}
>   ]
> {code}
> Screenshot -2
> This behavior should be documented.



--
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