[
https://issues.apache.org/jira/browse/HBASE-10273?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13861196#comment-13861196
]
Feng Honghua commented on HBASE-10273:
--------------------------------------
Two possibilities:
1. These buggy code has never been hit in production environment
2. Maybe some issues did arise due to these buggy code, but are ignored or
worked around , or nobody ever successfully dig from the symptom to these real
root cause. Actually *currently* HMaster does have much inconsistency, besides
the two places exposed by this jira, we don't have confidence that we have
figured out the root causes of all of them and fixed them, right? :-)
It's not every bug, especially of a large-scale distributed system such as
HBase, can be exposed by production running or covered by unit-test, until it
does show itself someday.
> AssignmentManager.regions(region to regionserver assignment map) and
> AssignmentManager.servers(regionserver to regions assignment map) are not
> always updated in tandem with each other
> ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-10273
> URL: https://issues.apache.org/jira/browse/HBASE-10273
> Project: HBase
> Issue Type: Bug
> Components: master
> Affects Versions: 0.94.16
> Reporter: Feng Honghua
> Assignee: Feng Honghua
> Fix For: 0.94.16
>
> Attachments: HBASE-10273-0.94_v0.patch
>
>
> By definition, AssignmentManager.servers and AssignmentManager.regions are
> tied and should be updated in tandem with each other under a lock on
> AssignmentManager.regions, but there are two places where this protocol is
> broken.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
