[
https://issues.apache.org/jira/browse/QPID-5715?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14012342#comment-14012342
]
Keith Wall edited comment on QPID-5715 at 5/29/14 1:16 PM:
-----------------------------------------------------------
Hi Alex,
Re. r1597801.
This is not as I understood how we wanted things to be. I thought, from the
*user* perspective (i.e. model upwards), the default replication policy
attributes would be:
localTransactionSyncronizationPolicy=SYNC
remoteTransactionSyncronizationPolicy=NO_SYNC
acknowledgmentPolicy=SIMPLE_MAJORITY
However, the implementation would be do things slightly differently
Local sync policy
=============
If the user chooses a local sync policy of SYNC, we are to implement this as JE
WRITE_NO_SYNC but with the coalescing committer to give the effect of SYNC but
in a more efficient manner.
If the user chooses a local sync policy of WRITE_NO_SYNC, normal JE rules would
apply. The transaction is written but the data would be committed to the disk
later. Coalescing committer would not be used. JE would perform the sync
later.
If the user chooses a local sync policy of NO_SYNC, normal JE rules would
apply. The transaction is neither written nor synced. Coalescing committer
would not be used. JE would perform the write/sync later.
Remote sync policy
===============
Normal JE rules apply.
I think we should expose acknowledgmentPolicy as far as the model (just in
case). It should default to SIMPLE_MAJORITY.
was (Author: k-wall):
Hi Alex,
Re. r1597801.
This is not as I understood how we wanted things to be. I thought, from the
*user* perspective (i.e. model upwards), the default replication policy
attributes would be:
localTransactionSyncronizationPolicy=SYNC
remoteTransactionSyncronizationPolicy=NO_SYNC
acknowledgmentPolicy=SIMPLE_MAJORITY
However, the implementation would be do things slightly differently
Local sync policy
=============
If the user chooses a local sync policy of SYNC, we are to implement this as JE
WRITE_NO_SYNC but with the coalescing committer to give the effect of SYNC but
in a more efficient manner.
If the user chooses a local sync policy of WRITE_NO_SYNC, normal JE rules would
apply. The transaction is written but the data would be committed to the disk
later. Coalescing committer would not be used. JE would perform the sync
later.
If the user chooses a local sync policy of NO_SYNC, normal JE rules would
apply. The transaction is neither written nor synced. Coalescing committer
would not be used. JE would perform the write/sync later.
Remote sync policy
===============
Normal JE rules apply.
I think we should expose replicaAcknowledgmentPolicy as far as the model (just
in case). It should default to SIMPLE_MAJORITY.
> [Java Broker] Introduce VirtualHostNode into model
> --------------------------------------------------
>
> Key: QPID-5715
> URL: https://issues.apache.org/jira/browse/QPID-5715
> Project: Qpid
> Issue Type: Sub-task
> Components: Java Broker
> Reporter: Keith Wall
> Assignee: Keith Wall
> Attachments:
> 0001-QPID-5715-Java-Broker-Make-virtualhosts-respect-the-.patch
>
>
> In order to support the HA use-case, a new object, VirtualHostNode will be
> added into the Broker model. The Broker will support zero or more virtual
> host nodes. Each virtual host node may have zero or one virtual hosts.
> Virtualhostnode will be responsible for
> * the lifecycle of the durable configuration store
> * the recovery of its children (that is, the virtual host and its object
> descendants (exchanges, queues etc) but not messages/message instances etc).
--
This message was sent by Atlassian JIRA
(v6.2#6252)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
