[
https://issues.apache.org/jira/browse/HBASE-18204?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16096934#comment-16096934
]
Xiaobing Zhou commented on HBASE-18204:
---------------------------------------
Here are some points:
# Not all exceptions from reading mean 'Connection closed to server', outgoing
RPCs shouldn't be treated as same by being cleaned as a result. Should we
respond based on different underlying errors?
# It actually introduces some bi-directional dependencies, e.g. RpcConnection
and ConnectionPool, RpcConnection and ClientDispatcher
# RpcConnection::SendRequest is not thread safe, e.g. when it comes to connect
> [C++] Rpc connection close and reconnecting
> -------------------------------------------
>
> Key: HBASE-18204
> URL: https://issues.apache.org/jira/browse/HBASE-18204
> Project: HBase
> Issue Type: Sub-task
> Reporter: Enis Soztutar
> Assignee: Enis Soztutar
> Attachments: hbase-18204_v1.patch, hbase-18204_v2.patch
>
>
> Our client-dispatcher maintains a map of outgoing RPCs per server with the
> promises.
> In case the server goes down, or TCP connection is closed, we should complete
> the outgoing RPCs with exceptions so that higher level waiters can unblock
> and retry.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
