gearpump-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Manu Zhang (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (GEARPUMP-31) Dynamic processor deletion of DAG
Date Thu, 14 Apr 2016 04:57:25 GMT

     [ https://issues.apache.org/jira/browse/GEARPUMP-31?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Manu Zhang updated GEARPUMP-31:
-------------------------------
    Component/s: streaming

> Dynamic processor deletion of DAG
> ---------------------------------
>
>                 Key: GEARPUMP-31
>                 URL: https://issues.apache.org/jira/browse/GEARPUMP-31
>             Project: Apache Gearpump
>          Issue Type: New Feature
>          Components: streaming
>            Reporter: JongHyok Lee
>
> Dynamic processor deletion of DAG required as a part of Dynamic DAG.
> Here's my usecase sample.
> First, let's assume that there is an application which is consisted with several processes
> (A) : Source which reads from Kafka
> (B) : Filter input with specific value
> (C) : Filter input with another specific value
> (D) : Sink input to HDFS
> (E) : Evaluate value of f(x1, x2, ...) where x1, x2, ... is part of input
> (F) : Transform input to another form
> (G) : Merge two input stream to one
> (H) : Sink input to Kafka topic
> (I) : Sink input to another Kafka topic
> (J) : Sink input to the third Kafka topic
> and the graph is
> (A) ~> (D)
> (A) ~> (B) ~> (E) ~> (H)
> (E) ~> (G) ~> (I)
> (A) ~> (C) ~> (F) ~> (J)
> (F) ~> (G)
> While the application is running, let's assume that there was a request from the consuming
component (it can be UI or other system which consumes kafka message) of the data stream that
they don't need merged data any more. Then maybe (G) ~> (I) part of the graph should be
removed from DAG. Also, if consuming component says they don't need different 'value' to filter,
then processor (C) should be removed and (F) should receive input from (F) (input might be
explicitly described in this case).
> This kind of functionality is really important for the system dealing with really huge
data streams, and so cannot let even part of DAG processors use CPU power when not needed.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message