socket = new DatagramSocket(12345);
byte[] buffer = new byte[1024];
DatagramPacket packet = new DatagramPacket(buffer, buffer.length);

Hi, I am currently concerned about the throughput of flink. I use flink socket stream. The current limit throughput of a single machine is 500,000 per second. I observe the CPU and memory of my machine, and their load is not high, so how to In order to improve the throughput of socket stream, where can I get the answer, thank you

There is no Flink DatagramSocket, isn't that Java itself?

Yes, I wrap my custom java data stream

Since it's not directly a Flink implementation, I don't think we can help much with that

The same is if I use the socketTextStream that comes with flink

Your source has a parallelism of 1

the source monitoring port 9999 obtains data, and the system prompts that this task cannot be parallelized

So there's the reason why you can't generate more data: it's a problem of your source

then for socketTextStream this will be a bottleneck ?

It’s a comparison of apples vs oranges

yes

But the feature set of both is completely different, so you can’t expect the same results.

Then wouldn't there be a better result?

Why?

for the Source task limited throughput

You’re running a flat map after a rebalance and a tumbling window after your source. The CPU and memory are not high. The metrics don’t show backpressure. From all indicators, your source isn’t generating enough data to produce more output.

Yep, the tumbling window is capturing its data according to your config. Most likely that means that it’s waiting for the window to close so it can work on the next set of data

is that mean waiting for the window to close cause a lot of data to be unprocessable?

Not unprocessable, but it’s basically waiting for the window to close before it can continue when a next window starts

then how to improve the job ?