[
https://issues.apache.org/jira/browse/HBASE-25007?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17194274#comment-17194274
]
Duo Zhang commented on HBASE-25007:
-----------------------------------
I think after a master restart, we will reload meta so we will know the regions
on the 'unknown servers', as we will create ServerStateNode for them.
And the problem is that, we will not schedule SCP for them if something goes
wrong. So I think a normal SCP is enough? Or at least, we do not need to scan
meta to find out the regions on a 'unknown server'?
At least, if meta is on an 'unknown server', the only way to fix is to restart
master, and then schedule a SCP manually for the region server. I do not think
schedule a HBCKSCP without restarting master could help here, as we will not
consider the region server as 'carryingMeta', so we will not try to assign meta.
In general, I think the HBCKChore should have two levels of scanning. First it
should always scan the location of meta(fetch zk for now, or scan root after
splittable meta, if we choose to use root table then there will be 3
levels...), and if there are inconsistencies, stop and report it. And once meta
is fine, we will scan meta to report other inconsistencies.
And if we find out that meta is already in inconsistent state, we schedule a
SCP for the region server, to bring meta online. And then the HBCKChore could
continue to show other inconsistencies, then we start a schedule other SCPs.
Notice that, we should restart master first to let the master load the newest
data in meta.
Thanks.
> Make HBCK2 work for 'root table'
> --------------------------------
>
> Key: HBASE-25007
> URL: https://issues.apache.org/jira/browse/HBASE-25007
> Project: HBase
> Issue Type: Sub-task
> Components: hbck2
> Reporter: Duo Zhang
> Assignee: Duo Zhang
> Priority: Major
>
> We will also scan catalog table and fix them in HBCK2, we should add support
> for root too.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
