[
https://issues.apache.org/jira/browse/SOLR-14273?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17044526#comment-17044526
]
Andrzej Bialecki commented on SOLR-14273:
-----------------------------------------
After discussing this issue with [~noble.paul] it turns out that the
{{#EQUALS}} condition that I used in this scenario behaves differently enough
that it may not need this attribute.
Namely, when {{#EQUALS}} is specified then it compares computed values per node
(in this case the number of cores) and records the amount of the deviation from
the ideal distribution.
For example, in case of 5 nodes and 7 replicas the {{#EQUALS}} condition can
never be satisfied exactly, because we can't put 1.4 replica on each node, we
can only put either 1 replica or 2 replicas. For the nodes where only 1 replica
is placed the deviation from the ideal state will be 0.4, and for the nodes
with 2 replicas the deviation will be -0.6. And as long as the {{abs(deviation)
< 1.0}} there's no violation, only potential optimizations are suggested.
> Autoscaling "cores" tag should support "strict : false"
> -------------------------------------------------------
>
> Key: SOLR-14273
> URL: https://issues.apache.org/jira/browse/SOLR-14273
> Project: Solr
> Issue Type: Improvement
> Security Level: Public(Default Security Level. Issues are Public)
> Components: AutoScaling
> Affects Versions: 8.4.1
> Reporter: Andrzej Bialecki
> Assignee: Andrzej Bialecki
> Priority: Major
> Fix For: 7.7.3, 8.5
>
>
> Currently when the "cores" global tag is parsed the parser automatically
> refuses to accept any other attributes - and since the "strict" attribute
> defaults to true there's no way to relax this rule.
> This becomes important when trying to spread the cores evenly across the
> cluster - even minor variations will generate violations, while in most cases
> a good enough spread is sufficient.
> It appears there's no other reason for this limitation apart from the rule
> parser limitation.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
