[
https://issues.apache.org/jira/browse/YARN-3367?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15070652#comment-15070652
]
Naganarasimha G R commented on YARN-3367:
-----------------------------------------
Thanks for the comments [~sjlee0],
bq. I would very much advocate using the JDK's ExecutorService (single-threaded
executor in this case) over using a raw thread and its own blocking queue
management.
May be i dint get your thoughts completely but let me explain the reason i have
taken this approach, Some points what i have considered :
* We require all the events to be pushed in the order with which its submitted.
Not sure whether we require order to be maintained across sync and async events
but definitely within sycn/ async its required based on the last [~djp]'s
comment. (for ex: metric events which are sent as async requires to be order
to ensure aggregation logic works properly and sync events like container
started / stopped so that state of it can be determined if there is any
intermittent daemon failures)
* As its single threaded better to merge the related events and push it once
(like all the waiting async events can be clubbed and pushed at once )
* for the Sync events we need to throw an exception on failure. so that caller
is informed that it failed.
Considering this i thought of maintaining a blocking queue and thread. so that
whenever the data is available then code in the thread can take some action,
(and by the time thread finishes publishing and comes back to read the queue if
multiple async entities are there it can merge and publish in next round.).
May be the complexity will get reduced if we *need not maintain the order*
across sync and async events.
Or please inform if i increased the scope of the jira than what is required.
bq. On TimelineEntities.java,
Yep can incorporate those changes, i just relied on eclipse auto code
generation for hash and equals for a given class.
> Replace starting a separate thread for post entity with event loop in
> TimelineClient
> ------------------------------------------------------------------------------------
>
> Key: YARN-3367
> URL: https://issues.apache.org/jira/browse/YARN-3367
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: timelineserver
> Affects Versions: YARN-2928
> Reporter: Junping Du
> Assignee: Naganarasimha G R
> Labels: yarn-2928-1st-milestone
> Attachments: YARN-3367-feature-YARN-2928.003.patch,
> YARN-3367-feature-YARN-2928.v1.002.patch,
> YARN-3367-feature-YARN-2928.v1.004.patch, YARN-3367.YARN-2928.001.patch
>
>
> Since YARN-3039, we add loop in TimelineClient to wait for
> collectorServiceAddress ready before posting any entity. In consumer of
> TimelineClient (like AM), we are starting a new thread for each call to get
> rid of potential deadlock in main thread. This way has at least 3 major
> defects:
> 1. The consumer need some additional code to wrap a thread before calling
> putEntities() in TimelineClient.
> 2. It cost many thread resources which is unnecessary.
> 3. The sequence of events could be out of order because each posting
> operation thread get out of waiting loop randomly.
> We should have something like event loop in TimelineClient side,
> putEntities() only put related entities into a queue of entities and a
> separated thread handle to deliver entities in queue to collector via REST
> call.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
