Hi,
i need to limit the rate of processing in a Flink stream application.
Specifically, the number of items processed in a .map() operation has to stay
under a certain maximum per second.
At the moment, I have another .map() operation before the actual processing,
which just sleeps for a certain time (e.g., 250ms for a limit of 4 requests /
sec) and returns the item unchanged:
…
public T map(final T value) throws Exception {
Thread.sleep(delay);
return value;
}
…
This works as expected, but is a rather crude approach. Checkpointing the job
takes a very long time: minutes for a state of a few kB, which for other jobs
is done in a few milliseconds. I assume that letting the whole thread sleep for
most of the time interferes with the checkpointing - not good!
Would using a different synchronization mechanism (e.g.,
https://google.github.io/guava/releases/19.0/api/docs/index.html?com/google/common/util/concurrent/RateLimiter.html)
help to make checkpointing work better?
Or, preferably, is there a mechanism inside Flink that I can use to accomplish
the desired rate limiting? I haven’t found anything in the docs.
Cheers,
Florian