[
https://issues.apache.org/jira/browse/BOOKKEEPER-208?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13284350#comment-13284350
]
Rakesh R commented on BOOKKEEPER-208:
-------------------------------------
Hi Ivan,
I had a quick look on the latest patch and the concept of write & ack quorum
looks nice to me.
Just I'm having following suggestion.
The BK client will maintains separate Write-Queue for each Bookie. What I'm
thinking is, quorumWrites will do parallely for each entries and it can be
asynchronously at the PendingAddOp level. Does it sound good to you?.
PendingAddOp.java
{noformat}
for (int bookieIndex : lh.distributionSchedule.getWriteSet(entryId)) {
sendWriteRequest(bookieIndex);
}
{noformat}
> Separate write quorum from ack quorum
> -------------------------------------
>
> Key: BOOKKEEPER-208
> URL: https://issues.apache.org/jira/browse/BOOKKEEPER-208
> Project: Bookkeeper
> Issue Type: New Feature
> Components: bookkeeper-client
> Reporter: Flavio Junqueira
> Fix For: 4.2.0
>
> Attachments: BOOKKEEPER-208.diff, BOOKKEEPER-208.diff
>
>
> There are use cases for bookkeeper that may require submitting add requests
> to a write set and returning upon receiving a confirmation from an ack set.
> The ack set must be a subset of the write set. An important special case is
> writing to all and returning upon hearing from a majority. Another important
> use case is avoiding *s* slow disks by writing to *f + s + 1* and returning
> upon receiving *f + 1* responses.
> Currently, the write set and the ack set are the same for a ledger. Internal
> changes to support these cases include changes to LedgerHandle and
> PendingAddOp. We also need to add a call to the client API to accept
> different sizes for the write set and the ack set upon ledger creation.
> It is also open for the discussion the need to implement a new distribution
> schedule. So far it looks like we can reuse the round robin implementation we
> currently have. We would need to implement a new one if, for example, the
> initial bookie of an add operation must be always the same.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
