[ 
https://issues.apache.org/jira/browse/HBASE-2315?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12846531#action_12846531
 ] 

Todd Lipcon commented on HBASE-2315:
------------------------------------

Thanks for attaching the design doc, Flavio. Here are a few questions generally 
about BK (would be great to address these in the docs too, not just this JIRA):

- Logs are stored replicated on several bookies. If a bookie goes down or loses 
a disk, the replication count obviously decreases. Will BK re-replicate the 
"under-replicated" parts of the log files? Or do we assume that logs are 
short-lived enough that we'll never lose all the replicas during the span of 
time for which the log is necessary?
- Similar to the above, is there any notion of rack awareness? eg does it 
ensure that each part of the log is replicated at least on two racks so that a 
full rack failure doesn't lose the logs?

Basically on a broad level what I'm asking is this: HDFS takes a lot of care to 
ensure availability of data "forever" across a variety of failure scenarios. 
From the design doc it wasn't clear that BK provides the same levels of safety. 
Can you please clarify what features (if any) are missing from BK that are 
present in HDFS with regard to reliability, etc?

> BookKeeper for write-ahead logging
> ----------------------------------
>
>                 Key: HBASE-2315
>                 URL: https://issues.apache.org/jira/browse/HBASE-2315
>             Project: Hadoop HBase
>          Issue Type: New Feature
>          Components: regionserver
>            Reporter: Flavio Paiva Junqueira
>         Attachments: bookkeeperOverview.pdf, HBASE-2315.patch, 
> zookeeper-dev-bookkeeper.jar
>
>
> BookKeeper, a contrib of the ZooKeeper project, is a fault tolerant and high 
> throughput write-ahead logging service. This issue provides an implementation 
> of write-ahead logging for hbase using BookKeeper. Apart from expected 
> throughput improvements, BookKeeper also has stronger durability guarantees 
> compared to the implementation currently used by hbase.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

Reply via email to