flink-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Chiwan Park <chiwanp...@apache.org>
Subject Re: Streaming in Flink
Date Tue, 05 Jan 2016 06:29:44 GMT
Hi Sourav,

Basically, Kafka consumer is pull-based [1]. If you want to build push-based system, you should
use other options.

Flink supports both pull-based and push-based paradigm. It depends upon an implementation
of data source. As one of examples, Flink provides a streaming source function based on socket

[1] http://kafka.apache.org/documentation.html#design_pull
[2] https://github.com/apache/flink/blob/master/flink-streaming-java/src/main/java/org/apache/flink/streaming/api/functions/source/SocketTextStreamFunction.java

> On Jan 5, 2016, at 3:01 PM, Sourav Mazumder <sourav.mazumder00@gmail.com> wrote:
> Hi Gordon,
> Need little more clarification around reading data from Kafka.
> As soon as any component behaves as a consumer of a topic/queue (iof a messaging system),
it essentially does polling of the data after a regular interval (that interval may be small
though). Hence essentially it captures all data/events accumulated in the queue from last
polling instant to the current polling in a pull manner
> This pattern is very different from real time push of data where a daemon process keeps
on waiting on a continuous basis for any data pushed to it. 
> So what I'm looking clarification for is whether Flink supports a mechanism where a data
source (actually any client application) can push a data to a socket which is continuously
listened by a deamon process of flink.
> Regards,
> Sourav
> On Mon, Jan 4, 2016 at 9:39 PM, Gordon Tai (戴資力) <tzulitai@gmail.com> wrote:
> Hi Sourav,
> Flink's streaming processes incoming data by-each-entry (true streaming, as compared
to micro-batch), and streaming is inherently designed as a push-model, where a topology of
stream transformations "listens" to a data source.
> You can have a Flink streaming topology's data source configured to be sockets or message
queues such as Kafka's topics.
> Any event / data that is streamed to (or in other words, "pushed" to) the socket or Kafka
topic will be processed by the Flink topology in real-time.
> Therefore, the answer is yes to your question. Hope this helps.
> BR,
> Gordon
> On Tue, Jan 5, 2016 at 12:42 PM, Sourav Mazumder <sourav.mazumder00@gmail.com>
> Hi,
> Does Flink support push based data streaming where the data source can push the events/data
to Flink cluster over a socket (instead of Flink pulling the data at a given frequency)?
> Regards,
> Sourav
> -- 
> Tzu-Li Tai (Gordon Tai)
> 戴資力
> National Cheng Kung University, Graduate Institute of Computer and Communication Engineering
> High Performance Parallel and Distributed Systems Laboratory (HPDS Lab)
> 國立成功大學電機工程學系 - 電腦與通信工程研究所
> 高效能平行/分散系統實驗室 (HPDS Lab)
> National Cheng Kung University, Engineering Science Dpt.
> 國立成功大學工程科學系
> Contacts
> tzulitai@ee.ncku.edu.tw
> http://tzulitai.ee.ncku.edu.tw
> Linkedin: tw.linkedin.com/in/tzulitai
> +886981916890

Chiwan Park

View raw message