[
https://issues.apache.org/jira/browse/IGNITE-18284?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Kirill Gusakov updated IGNITE-18284:
------------------------------------
Description:
According to the document attached in IGNITE-18060 we need to assign a separate
role, let's call it RebalanceScheduler. Under this issue we need to clarify:
* Which scope should we use for this role: cluster/table/partition-wide?
Depending on the answer to this question we can collocate this component with
placement driver/primary replica/partition group RAFT leader/CMG leader/leader
of metastorage group
* How will we guarantee, that only one scheduler per scope exists at the any
moment? Also, we must guarantee that there are no any stale operations from
previous scheduler overlap with the current one (by leasing techniques or
alternatives)
* How we will re-elect the new one, when current died.
was:
According to the document attached in IGNITE-18060 we need to assign a separate
role, let's call it RebalanceScheduler. Under this issue we need to clarify:
* Which scope should we use for this role: cluster/table/partition-wide?
Depending on the answer to this question we can collocate this component with
placement driver/primary replica/partition group RAFT laeder/CMG leader/leader
of metastorage group
* How will we guarantee, that only one scheduler per scope exists at the any
moment? Also, we must guarantee that there are no any stale operations from
previous scheduler overlap with the current one (by leasing techniques or
alternatives)
* How we will re-elect the new one, when current died.
> Prepare design for RebalanceScheduler role scope and election
> -------------------------------------------------------------
>
> Key: IGNITE-18284
> URL: https://issues.apache.org/jira/browse/IGNITE-18284
> Project: Ignite
> Issue Type: Task
> Reporter: Kirill Gusakov
> Assignee: Kirill Gusakov
> Priority: Major
> Labels: ignite-3
>
> According to the document attached in IGNITE-18060 we need to assign a
> separate role, let's call it RebalanceScheduler. Under this issue we need to
> clarify:
> * Which scope should we use for this role: cluster/table/partition-wide?
> Depending on the answer to this question we can collocate this component with
> placement driver/primary replica/partition group RAFT leader/CMG
> leader/leader of metastorage group
> * How will we guarantee, that only one scheduler per scope exists at the any
> moment? Also, we must guarantee that there are no any stale operations from
> previous scheduler overlap with the current one (by leasing techniques or
> alternatives)
> * How we will re-elect the new one, when current died.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
