murblanc commented on pull request #2285:
URL: https://github.com/apache/lucene-solr/pull/2285#issuecomment-773661420


   > Is not possible to make this a per collection feature? like the 
per-replica state feauture?
   
   Given Zookeeper content is identical for Overseer based and distributed 
cluster state update strategies (as opposed to the Per Replica States that use 
different znodes), a SolrCloud cluster can be switched between the two 
strategies by a single change of configuration. This makes it easy to test and 
adopt distributed updates, or to quickly rollback to Overseer based updates in 
case of issues.
   Doing this per collection would force modifying existing collections one by 
one.
   
   The following phase is distributing the Collection API execution. This 
distribution will rely on the version and conditional updates of `state.json` 
to replace Collection API locking. Dynamically changing how Collection API 
commands are executed for a given collection (Overseer or distributed) would be 
subject to tricky race conditions so would likely not be supported. Per Replica 
States collection configuration changes on the other hand are safe, guarded by 
the Collection API locking.
   
   In general, having all collections in the cluster follow the same execution 
and cluster state update strategies simplifies code and flow. For example 
enabling placement plugins to issue replica assignments for multiple 
collections or processing multiple collections for NodeDown.
   Also, changes to the way cluster state is cached on nodes will be simpler 
(and their impact on node and Zookeeper memory and CPU utilization observable) 
if the node maintains a single cache strategy. Otherwise, if any new cluster 
state cache is maintained in addition to the existing one (to serve both 
Overseer based and distributed Collection API calls), things will always look 
worse...


----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org



---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org

Reply via email to