[
https://issues.apache.org/jira/browse/IGNITE-18838?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Kirill Gusakov updated IGNITE-18838:
------------------------------------
Description:
According to the new replication design: distribution zones is the center of
data distribution, but not table partitions. So, the current logic for
start/stop needed RAFT nodes must be move to DistributionZoneManager (DZM).
*Defition of done:*
- DistributionZoneManager listens *.pending/*.stable assignment keys for all
distribution zones
- On update of pending key, DZM must start the needed raft group node
- On update of stable key, DZM must stop unneded node (so, the all nodes,
which doesn't belong to (pending ++ stable) set)
Important TODO: Also, we must to keep the logic, connected with the partition
listeners for new tables and index management.
was:
According to the new replication design: distribution zones is the center of
data distribution, but not table partitions. So, the current logic for
start/stop needed RAFT nodes must be move to DistributionZoneManager (DZM).
*Defition of done:*
- DistributionZoneManager listens \*.pending/\*.stable assignment keys for all
distribution zones
- On update of pending key, DZM must start the needed raft group node
- On update of stable key, DZM must stop unneded node (so, the all nodes,
which doesn't belong to (pending ++ stable) set)
> Start and stop of needed RAFT nodes must be moved from TableManager to
> DistributionZoneManager
> ----------------------------------------------------------------------------------------------
>
> Key: IGNITE-18838
> URL: https://issues.apache.org/jira/browse/IGNITE-18838
> Project: Ignite
> Issue Type: Task
> Reporter: Kirill Gusakov
> Priority: Major
> Labels: ignite-3
>
> According to the new replication design: distribution zones is the center of
> data distribution, but not table partitions. So, the current logic for
> start/stop needed RAFT nodes must be move to DistributionZoneManager (DZM).
> *Defition of done:*
> - DistributionZoneManager listens *.pending/*.stable assignment keys for all
> distribution zones
> - On update of pending key, DZM must start the needed raft group node
> - On update of stable key, DZM must stop unneded node (so, the all nodes,
> which doesn't belong to (pending ++ stable) set)
> Important TODO: Also, we must to keep the logic, connected with the partition
> listeners for new tables and index management.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
