[ 
https://issues.apache.org/jira/browse/BOOKKEEPER-592?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Sijie Guo updated BOOKKEEPER-592:
---------------------------------

    Attachment: BOOKKEEPER-592.patch

we are working on a rackaware ensemble placement, which is similar as the 
ticket here.

attach a patch of this work, so we could get some agreement on this part.

- abstract an EnsemblePlacementPolicy interface from BookieWatcher.
- separate the original ensemble select algorithm into 
DefaultEnsemblePlacementPolicy, which choose bookie randomly.
- provide a simple rack aware policy using hadoop's network topology library to 
build rack topology : choose the ensemble based on rack locality, which ensure 
there is at least two racks per write quorum. if this constraint could not be 
met, fallback to choose bookie randomly. the algorithm is simple and 
straightforward right now: pick up a bookie from the remote rack based on 
previous bookie to ensure there are at lease two racks per write quorum.


                
> allow application to recommend ledger data locality
> ---------------------------------------------------
>
>                 Key: BOOKKEEPER-592
>                 URL: https://issues.apache.org/jira/browse/BOOKKEEPER-592
>             Project: Bookkeeper
>          Issue Type: Improvement
>            Reporter: Ming Ma
>            Assignee: Sijie Guo
>         Attachments: BOOKKEEPER-592.patch
>
>
> For application like hbase WAL, it will be useful if application like hbase 
> can give a hint to bk about application's preferred ledger location. In that 
> way, application can fail over to specific machines where one of the ledger 
> replica is located; the recovery time will be faster. Another scenario is 
> hbase's support for hot standby region server where read request can be 
> served from a different machine other than the active region server. That 
> requires the hot standby region server to read from ledger. If the ledger is 
> on the same machine as standby region server, the performance will be better.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Reply via email to