hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "stack (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-9932) Remove Master Recovery handling when ZK session expired
Date Mon, 11 Nov 2013 21:54:18 GMT

    [ https://issues.apache.org/jira/browse/HBASE-9932?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13819463#comment-13819463
] 

stack commented on HBASE-9932:
------------------------------

[~jeffreyz] Go for it  We should do the "fail.fast.expired.active.master" by default.  The
'gray' area complicates more than it helps.   We already rehearsed the fail fast argument
over in "HBASE-9468 Previous active master can still serves RPC request when..."  We should
set it true in 0.98 and there remove code?  It may be too late for 0.96?

> Remove Master Recovery handling when ZK session expired
> -------------------------------------------------------
>
>                 Key: HBASE-9932
>                 URL: https://issues.apache.org/jira/browse/HBASE-9932
>             Project: HBase
>          Issue Type: Brainstorming
>            Reporter: Jeffrey Zhong
>
> Currently we use HMaster#tryRecoveringExpiredZKSession to allow master recovery from
a ZK session expired error. While this triggers to initialize HMaster partially, it is error
prone because it's hard to guarantee the half initialized master is in correct state. I found
several times already that the registered ZK listeners are different before & after a
fail over.
> Since we already have HA support, I'm proposing to remove this part handling. Though
we have a configuration setting "fail.fast.expired.active.master" to skip the logic, why not
go one stop further to clean the master code. 



--
This message was sent by Atlassian JIRA
(v6.1#6144)

Mime
View raw message