[
https://issues.apache.org/jira/browse/HBASE-23928?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17090855#comment-17090855
]
Michael Stack commented on HBASE-23928:
---------------------------------------
Yeah, currently fixing each Orphan means doing something like the below per
Orphan found:
{code}
$ hdfs dfs -lsr
hdfs://example.org/hbase/data/default/X/076d8d755631878da8935bc7e038bdce
$ hbase completebulkload
hdfs://example.org/hbase/data/default/X/076d8d755631878da8935bc7e038bdce X
$ hdfs dfs -lsr
hdfs://example.org/hbase/data/default/X/076d8d755631878da8935bc7e038bdce
$ hdfs dfs -rmr
hdfs://example.org/hbase/data/default/X/076d8d755631878da8935bc7e038bdce
{code}
The fixMeta in hbck2 does overlaps and holes. Perhaps add doing the above for
all regions found as part of the 'hbck2 filesystem --fix' command?
Of note, the completebulkload needs the table to be wholesome; should happen
AFTER the holes, inconsistences, and overlaps are fixed.
> Add hbck2 command to "completebulkload" a batch of orphaned regions
> -------------------------------------------------------------------
>
> Key: HBASE-23928
> URL: https://issues.apache.org/jira/browse/HBASE-23928
> Project: HBase
> Issue Type: Improvement
> Components: hbck2, Operability, Usability
> Reporter: Nick Dimiduk
> Assignee: Hua Xiang
> Priority: Major
>
> Cleaning up a cluster with a bunch of messed up region assignments and state,
> hbck2 gets us to the point where meta and hdfs agree on split points, but
> some regions are left orphaned in hdfs. The current solution is to manually
> loop through the list of orphans and invoke {{completebulkload}} one-by-one.
> Let's add this as an action that hbck2 and do on the operator's behalf.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
