Hi!

I just tracked down why kafka was using excessive CPU when storm topologies
were deployed.  It turns out that this setting:

topology.sleep.spout.wait.strategy.time.ms

when left at the default (1 ms) pegs kafka's CPU.  I read the FAQ (I think,
I read a FAQ at least) that briefly mentions the delay.  Certainly changing
to something like, say, 1000 makes kafka's CPU go back down when idle.

So, just a general question about this setting?  Is this what people use in
production environments?  Are there better wait strategies for the kafka
spout?  What's the motivation for having the default be 1 ms?  Any other
advice?

Thanks!

Cheers,
Craig

Reply via email to