[
https://issues.apache.org/jira/browse/HBASE-15331?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15191388#comment-15191388
]
Vladimir Rodionov commented on HBASE-15331:
-------------------------------------------
Partially resolved (in HBASE-14123-v11 patch).
Now we restore all intermediate incremental images in a single M/R job instead
of multiple (one per image) and we use bulk load and nor streaming puts.
> HBase Backup/Restore Phase 2: Optimized Restore operation
> ---------------------------------------------------------
>
> Key: HBASE-15331
> URL: https://issues.apache.org/jira/browse/HBASE-15331
> Project: HBase
> Issue Type: Improvement
> Reporter: Vladimir Rodionov
> Assignee: Vladimir Rodionov
>
> The current implementation for restore uses WALReplay M/R job. This has
> performance and stability problems, since it uses HBase client API to insert
> data. We have to migrate to bulk load approach: generate hfiles directly from
> snapshot and incremental images. We run separate M/R job for every backup
> image between last FULL backup and current incremental backup we restore to
> and for every table in the list (image). If we have 10 tables and 30 days of
> incremental backup images - this results in 30x10 = 300 M/R jobs. MUST be
> optimized.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
