[
https://issues.apache.org/jira/browse/OAK-4083?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15318989#comment-15318989
]
Alex Parvulescu commented on OAK-4083:
--------------------------------------
merged to 1.2 with r1747284
> Simplify concurrency when loading data from the primary
> -------------------------------------------------------
>
> Key: OAK-4083
> URL: https://issues.apache.org/jira/browse/OAK-4083
> Project: Jackrabbit Oak
> Issue Type: Bug
> Components: tarmk-standby
> Reporter: Francesco Mari
> Assignee: Francesco Mari
> Fix For: 1.4.1, 1.5.0, 1.2.16
>
> Attachments: OAK-4083-01.patch
>
>
> The cold standby client is overly complicated and potentially buggy when
> segments are loaded from the primary.
> In particular:
> - when the {{StandbyClientHandler}} is removed from the pipeline, its
> {{EventExecutorGroup}} is passed to the {{SegmentLoaderHandler}} to signal a
> surrogate of the "channel close" event. It would be more sensible for
> {{SegmentLoaderHandler}} to listen to the proper "channel close" event
> instead.
> - after the {{SegmentLoaderHandler}} is added to the pipeline, the "channel
> active" callback method is called directly using the
> {{ChannelHandlerContext}} of the {{StandbyClientHandler}}. It would be better
> for {{SegmentLoaderHandler}} to use a proper initialisation point like, in
> example, the "handler added" callback method. The {{SegmentLoaderHandler}}
> should also use its own {{ChannelHandlerContext}}, instead of owning one from
> another handler.
> - {{SegmentLoaderHandler}} saves and uses the {{ChannelHandlerContext}} of
> the {{StandbyClientHandler}}. The only purpose for it is to use the
> {{EventExecutorGroup}} of the {{StandbyClientHandler}} to issue "write
> messages" events. The {{SegmentLoaderHandler}} should use Netty's event loop
> to issue I/O operations, and instead use a different thread for business
> logic - in this case, the synchronization process between primary and standby
> instances.
> - The {{StandbyClientHandler}} is registered with its own
> {{EventExecutorGroup}} with four threads. This is overkill, since the
> synchronization process should run anyway in a separate thread (see below).
> The {{StandbyClientHandler}} should use the default event loop.
> - The {{SegmentLoaderHandler}} is registered with its own
> {{EventExecutorGroup}} with four threads. This is overkill, since
> request/response cycle between the standby and the primary instances is
> synchronous. The {{SegmentLoaderHandler}} should instead use the default
> event loop for I/O events and run another (single) thread to execute the
> synchronization process.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
