hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Yong Zhang (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-15318) Zk-less region server state management
Date Wed, 24 Feb 2016 16:48:18 GMT

    [ https://issues.apache.org/jira/browse/HBASE-15318?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15163304#comment-15163304
] 

Yong Zhang commented on HBASE-15318:
------------------------------------

Thanks [~mbertozzi] for explain.
bq. So, the ephemeral nodes the RS registers on startup
Yes, RS will send a heartbeat to HM.

bq. When you say ..."but we find many user network issue is not network disconnected but package
lost", is it that the RS is dead/lost (but its zk connection is fine?)
Here just describe one case, that when RS start, it will create one znode on zk, but some
time later, network package has 10% lost for example, because connection from this RS to ZK
is not break then, HM also consider this RS is health, but in fact this RS may could not provide
service.

> Zk-less region server state management
> --------------------------------------
>
>                 Key: HBASE-15318
>                 URL: https://issues.apache.org/jira/browse/HBASE-15318
>             Project: HBase
>          Issue Type: Improvement
>            Reporter: Yong Zhang
>            Assignee: Yong Zhang
>
> Current region server state is managed via znode created by region server, master just
listen these nodes. but we find many user network issue is not network disconnected but package
lost, which is hard to capture because connection between region server and zk is fine.
> This jira goal is region server state is managed by master without shared info in zk,
via enhancement heartbeat from region server to master.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message