[
https://issues.apache.org/jira/browse/IGNITE-23805?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Denis Chudov updated IGNITE-23805:
----------------------------------
Description:
Design the changes related to nodes' joins and leaves. If the node matches
zone's filters and storage profiles, then the zone should be scaled up/down
according to its scale up/down timers:
* on scale up, the learner should be automatically added, if there are enough
peers, otherwise peer should be added;
* on scale down, the leave of a learner probably doesn't cause any actions
(check it), the leave of a peer should lead to transformation of a learner into
peer. JRaft built-in features should be considered to do this;
* defaults of scale up/down timers for replicated zones may be reconsidered.
Note: there is a possibility that we will need to limit the learners count as
well (by limiting the total number of replicas).
Note 2: this part of design touches other parts, especially peer/learner
assignments and rebalancing.
> Design scale up/scale down for replicated zones
> -----------------------------------------------
>
> Key: IGNITE-23805
> URL: https://issues.apache.org/jira/browse/IGNITE-23805
> Project: Ignite
> Issue Type: Task
> Reporter: Denis Chudov
> Priority: Major
> Labels: ignite-3
>
> Design the changes related to nodes' joins and leaves. If the node matches
> zone's filters and storage profiles, then the zone should be scaled up/down
> according to its scale up/down timers:
> * on scale up, the learner should be automatically added, if there are
> enough peers, otherwise peer should be added;
> * on scale down, the leave of a learner probably doesn't cause any actions
> (check it), the leave of a peer should lead to transformation of a learner
> into peer. JRaft built-in features should be considered to do this;
> * defaults of scale up/down timers for replicated zones may be reconsidered.
> Note: there is a possibility that we will need to limit the learners count as
> well (by limiting the total number of replicas).
> Note 2: this part of design touches other parts, especially peer/learner
> assignments and rebalancing.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
