[
https://issues.apache.org/jira/browse/HBASE-24590?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17141272#comment-17141272
]
Bharath Vissapragada commented on HBASE-24590:
----------------------------------------------
> That way not only we require any additional step in CloneSnapshotProcedure,
> but also the responsibility of overriding attribute remains with
> SnapshotManager.
The reason is that some "modifications" need more than updating the
TableDescriptor in place. For example, if someone changes region replica count,
we also need to add/remove additional replicas before enabling the table. This
is all handled neatly by ModifyTableProcedure.
> Clone snapshot with optional table level property overrides
> -----------------------------------------------------------
>
> Key: HBASE-24590
> URL: https://issues.apache.org/jira/browse/HBASE-24590
> Project: HBase
> Issue Type: Improvement
> Components: API, master, shell
> Affects Versions: 3.0.0-alpha-1, 2.3.0, 1.6.0
> Reporter: Bharath Vissapragada
> Priority: Minor
>
> We have this use case where we try to clone from snapshots of a table with
> replication enabled. Cloning process just reads the TableDescriptor from the
> manifest and uses it for the new target table.
> In our case, the moment we clone it, replication kicks in and attempts to
> replicate table to peers but since the target table doesn't exist, it keeps
> throwing TNFE in a loop. I propose to add additional *optional* parameters to
> clone_snapshot that takes the target table property overrides (table leve/cf
> level).
> For example:
> {noformat}
> clone_snapshot 'snapshotName', 'tableName' # default
> clone_snapshot 'snapshotName', 'tableName', {NAME => 'f1', REPLICATION_SCOPE
> => 0 }
> clone_snapshot 'snapshotName, 'tableName', {MERGE_ENABLED => false}
> .....
> {noformat}
>
> Implementation wise, this just adds a {{ModifyTableProcedure}} as a child
> procedure (if args are provided) for {{CloneSnapshotProcedure}} after
> {{CLONE_SNAPSHOT_ADD_TO_META}}, so we reuse most of the "modifyTable" code.
> Also, since clone puts the table in a disabled state, using
> "ModifyTableProcedure" shouldn't cause any issues.
> Thoughts? Is there any easier way to do it?
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
