[
https://issues.apache.org/jira/browse/HBASE-20788?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16523261#comment-16523261
]
Duo Zhang commented on HBASE-20788:
-----------------------------------
{quote}
its this true? What if the 2.0 server carrying hbase:meta goes down before
another hbase2 is up on the cluster?
{quote}
Usually it will be fine. As we will upgrade master at the end, so no one will
write to the new table state family. But if you have tables which have
REPLICATION_SCOPE set to 1, then for 2.1.0 there will be problem. In 2.1.0,
when opening a region with REPLICATION_SCOPE set to 1, we will write a barrier
in the new replication barrier family, for serial replication. So if the meta
region is then assigned to a RS with the old version, then it will fail to
replay the wal and meta region could never online. For 2.0 there is no problem.
But anyway, for upgrading you will soon have several RSes with the new
version... And it is known that we do not support downgrading, so...
> Write up a doc about how to rolling upgrade from 1.x to 2.x
> -----------------------------------------------------------
>
> Key: HBASE-20788
> URL: https://issues.apache.org/jira/browse/HBASE-20788
> Project: HBase
> Issue Type: Sub-task
> Components: documentation
> Reporter: Duo Zhang
> Assignee: Duo Zhang
> Priority: Blocker
> Fix For: 2.1.0
>
> Attachments: HBASE-20788.patch
>
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
