[
https://issues.apache.org/jira/browse/HBASE-26220?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17419301#comment-17419301
]
Duo Zhang edited comment on HBASE-26220 at 9/23/21, 4:19 PM:
-------------------------------------------------------------
I think this is a very important improvements.
In the current implementation, every region server needs to watch all the
region servers in the cluster, through a zk watcher, which put big pressure on
zookeeper for a large cluster.
We'd better land this before cutting 2.5.0, but maybe we need a backport issue
as the implementation will be different since there is no
AsyncClusterConnection for branch-2.
was (Author: apache9):
I think this is a very important improvements.
In the current implementation, every region server needs to watch all the
region servers in the cluster, through a zk watcher, which put big pressure on
zookeeper for a large cluster.
We'd better land this before cutting 2.5.0.
> Use P2P communicate between region servers to sync the list for bootstrap node
> ------------------------------------------------------------------------------
>
> Key: HBASE-26220
> URL: https://issues.apache.org/jira/browse/HBASE-26220
> Project: HBase
> Issue Type: Sub-task
> Reporter: Duo Zhang
> Assignee: Duo Zhang
> Priority: Major
>
> For returning bootstrap nodes to client, we do not need to always track
> exactly the current live region server. But now, we just set a watcher on
> zookeeper to watch all the live region servers, on every region server, for
> large cluster, it will be a big pressure for zookeeper.
> So here I think, we could let region servers to communicate with each other
> to sync the available bootstrap node, like a P2P network.
> Will put a simple design doc soon.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
