[
https://issues.apache.org/jira/browse/HBASE-20065?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16376007#comment-16376007
]
Duo Zhang commented on HBASE-20065:
-----------------------------------
I always wonder why we need to set timestamp explicitly when updating meta
region.
But if we have a bad clock then only something like HLC can save us... For
example, if you use the local time for master as timestamp, then a new master
with a smaller time(no sure how to describe this in English...) will kill you.
But if you use HConstants.LATEST_TIMESTAMP, which meas you want to use the
local time of the RS which holds meta, then if meta is assigned to another RS
with a smaller time will kill you...
{quote}
HBCK doesn't work against hbase2 and setting time on client-side seems
problematic to me.
{quote}
[~openinx] said above that it is a little confusing that we use current time
everywhere but HConstants.LATEST_TIMESTAMP only in hbck. For me I'm OK with
both approach...
Thanks.
> Revisit the timestamp usage in MetaTableAccessor
> ------------------------------------------------
>
> Key: HBASE-20065
> URL: https://issues.apache.org/jira/browse/HBASE-20065
> Project: HBase
> Issue Type: Improvement
> Reporter: Duo Zhang
> Assignee: Duo Zhang
> Priority: Major
> Fix For: 2.0.0-beta-2
>
> Attachments: HBASE-20065-v1.patch, HBASE-20065.patch
>
>
> It is totally a mess and makes me confusing when reimplementing the serial
> replication feature. Let me do a clean up first.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
