[
https://issues.apache.org/jira/browse/IGNITE-19443?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Mirza Aliev updated IGNITE-19443:
---------------------------------
Description:
*Motivation:*
After https://issues.apache.org/jira/browse/IGNITE-18955 is implemented, it is
possible, that data nodes will be empty according to provided filter. Such data
nodes should not be propagated to pending assignments (meaning that new
rebalance should not be triggered)
*Definition of done:*
* We have two places, where such propagation can be done, one is a reaction on
replica factor hanging, another one is when data nodes was changed. Both places
must skip pending key propagation if data nodes become empty after filtering.
* Corresponding tests for both scenarios must be provided
was:
*Motivation: *
After https://issues.apache.org/jira/browse/IGNITE-18955 is implemented, it is
possible, that data nodes will be empty according to provided filter. Such data
nodes should not be propagated to pending assignments (meaning that new
rebalance should not be triggered)
*Definition of done:*
* We have two places, where such propagation can be done, one is a reaction on
replica factor hanging, another one is when data nodes was changed. Both places
must skip pending key propagation if data nodes become empty after filtering.
* Corresponding tests for both scenarios must be provided
> Empty data nodes shouldn't be propagated to pending assignments after
> filtering
> ---------------------------------------------------------------------------------
>
> Key: IGNITE-19443
> URL: https://issues.apache.org/jira/browse/IGNITE-19443
> Project: Ignite
> Issue Type: Bug
> Reporter: Mirza Aliev
> Priority: Major
> Labels: ignite-3
>
> *Motivation:*
> After https://issues.apache.org/jira/browse/IGNITE-18955 is implemented, it
> is possible, that data nodes will be empty according to provided filter. Such
> data nodes should not be propagated to pending assignments (meaning that new
> rebalance should not be triggered)
> *Definition of done:*
> * We have two places, where such propagation can be done, one is a reaction
> on replica factor hanging, another one is when data nodes was changed. Both
> places must skip pending key propagation if data nodes become empty after
> filtering.
> * Corresponding tests for both scenarios must be provided
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
