[
https://issues.apache.org/jira/browse/HBASE-17123?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16277367#comment-16277367
]
Appy commented on HBASE-17123:
------------------------------
you're right, bulk loading is over, so maybe false cannot be interpreted as
failure.
But we still need to keep backward compat, so best is documenting the behavior.
Maybe mention the meaning of return value in javadoc of observer (that
bulkload is done by post*() call and false doesn't rollback/cancel it) and
proto (that false is ambiguous in case of CPs, can mean failed op, or
otherwise) in your HBASE-19417 fix [~tedyu]?
> Add postBulkLoadHFile variant that notifies the final paths for the hfiles
> --------------------------------------------------------------------------
>
> Key: HBASE-17123
> URL: https://issues.apache.org/jira/browse/HBASE-17123
> Project: HBase
> Issue Type: Improvement
> Reporter: Ted Yu
> Assignee: Ted Yu
> Labels: bulkloader
> Fix For: 2.0.0
>
> Attachments: 17123.addendum, 17123.v1.txt, 17123.v3.txt,
> 17123.v4.txt, 17123.v5.txt
>
>
> Currently the postBulkLoadHFile() hook passes the same familyPaths parameter
> which it receives as method parameter.
> See code in SecureBulkLoadManager :
> {code}
> loaded = region.getCoprocessorHost().postBulkLoadHFile(familyPaths,
> loaded);
> {code}
> Meaning, the paths are not final, moved paths of the loaded hfiles.
> This issue is to add a variant which notifies the final paths of the loaded
> hfiles.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
