[
https://issues.apache.org/jira/browse/SOLR-12847?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17083310#comment-17083310
]
Andrzej Bialecki edited comment on SOLR-12847 at 4/14/20, 3:07 PM:
-------------------------------------------------------------------
I started working on this (see the PR) but quite a few tests started failing
when this automatically-created policy has been added instead of
{{maxShardsPerNode}}.
It turns out that in many cases the auto-created policy overrode the cluster
policy, specifically the non-strict rule that specifies the max number of
replicas per node. It seems that using just the autoscaling rules it's not
possible in general case to both preserve the existing semantics of
{{maxShardsPerNode}} and also preserve the cluster policy rules - because the
auto-created policy will override the cluster policy.
In any case, in branch_8x the only thing that {{maxShardsPerNode}} does is to
simply compare this value with an average number of collection replicas per
node, completely ignoring rules, so it's useful only as a crude safety guard
anyway.
For this reason, in order to preserve back-compat I'd like to leave the code
as-is on branch 8x and instead completely remove {{maxShardsPerNode}} from
master (9.0).
was (Author: ab):
I started working on this (see the PR) but quite a few tests started failing
when this automatically-created policy has been added instead of
{{maxShardsPerNode}}.
It turns out that in many cases the auto-created policy overrode the cluster
policy, specifically the non-strict rule that specifies the max number of
replicas per node. It seems that using just the autoscaling rules it's not
possible in general case to both preserve the existing semantics of
{{maxShardsPerNode}} and also preserve the cluster policy rules - because the
auto-created policy will override the cluster policy.
For this reason, in order to preserve back-compat I'd like to leave the code
as-is on branch 8x and instead completely remove {{maxShardsPerNode}} from
master (9.0).
> Cut over implementation of maxShardsPerNode to a collection policy
> ------------------------------------------------------------------
>
> Key: SOLR-12847
> URL: https://issues.apache.org/jira/browse/SOLR-12847
> Project: Solr
> Issue Type: Bug
> Components: AutoScaling, SolrCloud
> Reporter: Shalin Shekhar Mangar
> Assignee: Andrzej Bialecki
> Priority: Major
> Fix For: master (9.0)
>
> Time Spent: 0.5h
> Remaining Estimate: 0h
>
> We've back and forth over handling maxShardsPerNode with autoscaling policies
> (see SOLR-11005 for history). Now that we've reimplemented support for
> creating collections with maxShardsPerNode when autoscaling policy is
> enabled, we should re-look at how it is implemented.
> I propose that we fold maxShardsPerNode (if specified) to a collection level
> policy that overrides the corresponding default in cluster policy (see
> SOLR-12845). We'll need to ensure that if maxShardsPerNode is specified then
> the user sees neither violations nor corresponding suggestions because of the
> default cluster policy.
--
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
