[
https://issues.apache.org/jira/browse/BOOKKEEPER-325?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13450666#comment-13450666
]
Ivan Kelly commented on BOOKKEEPER-325:
---------------------------------------
new patch looks good. There's a few minor things.
- better to call config option openLedgerRereplicationGracePeriod, Should be in
ServerConfiguration, not AbstractConfiguration. The javadoc needs more detail.
{code}
/**
* Set the grace period which the rereplication worker will wait before
* fencing and rereplicating a ledger fragment which is still being written
* to, on bookie failure.
*
* The grace period allows the writer to detect the bookie failure, and
* and start writing to another ledger fragment. If the writer writes nothing
* during the grace period, the rereplication worker assumes that it has crashed
* and therefore fences the ledger, preventing any further writes to that
ledger.
*
* @see LedgerHandle#openLedger
*/
{code}
- getUnderreplicatedFragments should only take a LedgerHandle. It can construct
the CheckerCallback itself.
- rename (releaseDeferredLedgerLock -> deferLedgerLockRelease)
- indentation on first line of releaseDeferredLedgerLock is off.
> Delay the replication of a ledger if RW found that its last fragment is in
> underReplication.
> --------------------------------------------------------------------------------------------
>
> Key: BOOKKEEPER-325
> URL: https://issues.apache.org/jira/browse/BOOKKEEPER-325
> Project: Bookkeeper
> Issue Type: Sub-task
> Components: bookkeeper-auto-recovery
> Affects Versions: 4.2.0
> Reporter: Uma Maheswara Rao G
> Assignee: Uma Maheswara Rao G
> Attachments: BOOKKEEPER-325.patch, BOOKKEEPER-325.patch
>
>
> When RW found that ledger's last fragment is in underReplication state, then
> we should delay that ledger replication for some grace period. optimally we
> can replicate other fragments.
> The idea is, Whenever it finds the last fragement is under replicated, It can
> add into PendingReplication list.
> There will be a small daemon, which will check for the timeouts of this
> ledgers.
> Once it timed out , it will trigger the normal replication process if it is
> not in last fragment. Otherwise, it will fence the ledger and will trigger
> the replication nomally.
> see the discussion for more info:
> http://markmail.org/message/ruhhxxgvuqnjlu2s#query:+page:1+mid:f6ifo4sizulwiaem+state:results
--
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
