Re: How to broadcast messages to all task manager instances in cluster?

Posted by Di Tang on
URL: http://deprecated-apache-flink-user-mailing-list-archive.369.s1.nabble.com/How-to-broadcast-messages-to-all-task-manager-instances-in-cluster-tp20091p20111.html

Thanks Piotr for the response. I have many data streams dependant on the configuration by getting value from static variables in a class. The way the configuration change works is to change the static variables' value in the class. Since each task manager only has one JVM process, as long as the message is broadcast to each task manager, the data streams will see the change. The logic in data streams is quite simple, just get some parameters from the static variable. So I think to add connect and flatmap to each of them is too verbose. I am wondering is there any better way to express.

Piotr Nowojski <[hidden email]> 于 2018年5月11日周五 下午7:31写道：

Hi,

I don’t quite understand your problem. If you broadcast message as an input to your operator that depends on this configuration, each instance of your operator will receive this configuration. It shouldn't matter whether Flink scheduled your operator on one, some or all of the TaskManagers. It only should matter if operators running your configuration sensitive code receive the broadcasted message.

DataStream<> input = xxx;

DataStream<> controlConfigInput = yyy;

DataStream<> data = input.

.do()

.something()

.fancy();

controlConfigInput.broadcast()

.connect(data)

.flatMap(new MyFancyOperatorThatDependsOnConfigStream())

Or slide 36 from here: https://www.slideshare.net/dataArtisans/apache-flink-datastream-api-basics

Piotrek

On 11 May 2018, at 11:11, Di Tang <[hidden email]> wrote:

Hi guys:

I have a Flink job which contains multiple pipelines. Each pipeline depends on some configuration. I want to make the configuration dynamic and effective after change so I created a data source which periodically poll the database storing the configuration. However, how can I broadcast the events to all task manager instances?  The datastream.broadcast() only applies to the parallel instances of operator. And I don't want to connect the configuration data source to each pipeline because it is too verbose. If Flink cannot explicitly broadcast messages to task managers, is there any method to guarantee the parallel operator is distributed on all task managers?

Thanks,

Di