[
https://issues.apache.org/jira/browse/HBASE-11742?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14102809#comment-14102809
]
Andrew Purtell commented on HBASE-11742:
----------------------------------------
What happens if there is an older client library that wants to run a MR-over
snapshots job against snapshots dropped by newer servers? The older library
will be looking for table descriptors and snapshot region names in the FS
instead of SnapshotFileInfo/SnapshotRegionManifest.
Do we continue to handle TableSnapshotRegionSplit messages that have
RegionSpecifiers (field #1) instead of the new TableSchema and RegionInfo
fields (3 and 4)?
What happens if someone uses the newer client in a MR-over-snapshots job where
there is no SnapshotFileInfo/SnapshotRegionManifest data available because the
servers are older?
What happens when we have a mix of snapshots dropped by an older server side
versus a newer server side?
> Backport HBASE-7987 and HBASE-11185 to 0.98
> -------------------------------------------
>
> Key: HBASE-11742
> URL: https://issues.apache.org/jira/browse/HBASE-11742
> Project: HBase
> Issue Type: Improvement
> Components: mapreduce, snapshots
> Affects Versions: 0.98.5
> Reporter: Esteban Gutierrez
> Assignee: Esteban Gutierrez
> Fix For: 0.98.6
>
> Attachments: HBASE-11742.v0.patch, HBASE-11742.v1.patch
>
>
> HBASE-7987 improves how snapshots are handled via a manifest file. This
> requires reverting HBASE-11360 since introduces an alternate functionality
> that is not compatible with HBASE-7987.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
