[
https://issues.apache.org/jira/browse/HBASE-19623?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16305420#comment-16305420
]
Duo Zhang commented on HBASE-19623:
-----------------------------------
Skimmed the code of ReplicationSourceManager, ReplicationSource and
ReplicationSourceShipper, I think we can move the initialization of
ReplicationEndpoint to the run method of ReplicationSource. We will try
initializing the ReplicationEndpoint at the beginning and if it is failed we
will retry forever, until the replication source is terminated, i.e., removed.
I think this could solve the problem.
> Create the connection to peer cluster async when region server add a
> replication source.
> -----------------------------------------------------------------------------------------
>
> Key: HBASE-19623
> URL: https://issues.apache.org/jira/browse/HBASE-19623
> Project: HBase
> Issue Type: Sub-task
> Components: proc-v2, Replication
> Reporter: Zheng Hu
> Assignee: Zheng Hu
>
> As the discussion in HBASE-19617, After the replication procedure replace
> the zookeeper notification , the addPeer operation may be blocked because
> the RegionServer will create a connection to peer cluster synchronously.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
