[
https://issues.apache.org/jira/browse/HDDS-10714?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17838804#comment-17838804
]
Ethan Rose commented on HDDS-10714:
-----------------------------------
Here's what I'm thinking so far. Here a *target* is a node that receives a
reconcile command, and a *peer* is a node given to the target to contact for
reconciliation information.
* Stale and dead nodes are not considered as peers or targets.
* Decommissioned or Maintenance mode nodes are not considered as peers or
targets.
** These are considered to have left the cluster already. If they return with a
different hash, another round of reconciliation can be run.
* Decommissioning nodes can be used as peers but will not be targets.
** We want to make sure we have moved all data off of these nodes. This will be
a pre-requisite of decommissioning when reconciliation is integrated with the
replication manager.
* Entering maintenance nodes can be used as peers but will not be targets.
** Maintenance guarantees availability, not durability. If the entering
maintenance node has a unique hash, it may need to be a peer for reconciliation
on another node to ensure it doesn't have unique data that will become
unreadable when it moves to maintenance mode.
* All other nodes are eligible to be peers and targets.
> Restrict reconciliation requests by datanode status
> ---------------------------------------------------
>
> Key: HDDS-10714
> URL: https://issues.apache.org/jira/browse/HDDS-10714
> Project: Apache Ozone
> Issue Type: Sub-task
> Reporter: Ethan Rose
> Priority: Major
>
> HDDS-10372 allows reconciliation requests to go to all nodes with a
> container, and all nodes to be used as peers. We should make some
> restrictions based on node state (healthy, stale, dead) and operational state
> (decommission/maintenance).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
