[
https://issues.apache.org/jira/browse/SOLR-17604?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
David Smiley updated SOLR-17604:
--------------------------------
Description:
This is a refactoring but not a hidden one and it impacts performance
positively for both ZK and HTTP ClusterStateProviders.
Propose that collection state caching responsibilities move from
CloudSolrClient to the ClusterStateProvider. CSC needs to obtain a
DocCollection from CSP (exists) and needs to inform CSP on occasion that
there's a version of a DocCollection at a higher version (new method –
informNewVersion(collName,collVersion), so that the CSP can invalidate/update
it's cache accordingly (details TBD and may evolve). The ZK CSP would do
nothing since it watches everything (for better or worse) and it'll get up to
date on its own. Also, remove ClusterState.CollectionRef concept from CSP's
API, thus remove the getState method.
Unclear if alias resolution methods need work.
was:
This is a refactoring but not a hidden one and it impacts performance
positively for both ZK and HTTP ClusterStateProviders.
Propose that collection state caching responsibilities move from
CloudSolrClient to the ClusterStateProvider. CSC needs to obtain a
DocCollection from CSP (exists) and needs to inform CSP on occasion that
there's a version of a DocCollection at a higher version (new method –
informNewVersion(collName,collVersion), so that the CSP can invalidate/update
it's cache accordingly (details TBD and may evolve). The ZK CSP would do
nothing since it watches everything (for better or worse) and it'll get up to
date on its own. Also, hopefully can remove ClusterState.CollectionRef
concept from CSP's API.
Separately: deprecating ClusterState access altogether by CSC & CSP (albeit
will exist specifically on the ZK CSP). CSP can have methods to obtain
whatever someone may want and/or users can use standard Solr admin APIs to
obtain info. Probably needs a collection-list method.
> Move state caching from CloudSolrClient to ClusterStateProvider
> ---------------------------------------------------------------
>
> Key: SOLR-17604
> URL: https://issues.apache.org/jira/browse/SOLR-17604
> Project: Solr
> Issue Type: Improvement
> Components: SolrCloud, SolrJ
> Reporter: David Smiley
> Priority: Major
>
> This is a refactoring but not a hidden one and it impacts performance
> positively for both ZK and HTTP ClusterStateProviders.
> Propose that collection state caching responsibilities move from
> CloudSolrClient to the ClusterStateProvider. CSC needs to obtain a
> DocCollection from CSP (exists) and needs to inform CSP on occasion that
> there's a version of a DocCollection at a higher version (new method –
> informNewVersion(collName,collVersion), so that the CSP can invalidate/update
> it's cache accordingly (details TBD and may evolve). The ZK CSP would do
> nothing since it watches everything (for better or worse) and it'll get up to
> date on its own. Also, remove ClusterState.CollectionRef concept from CSP's
> API, thus remove the getState method.
> Unclear if alias resolution methods need work.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
