Thanks Enno. I tried these days to play with some Storm parameters (MaxParalelism) and settings (running multiple tasks for a busy Bolt), but still observing same pattern. I also stopped Kafka spout (i.e. just no sending any more data to Kafka) and monitored the busy Bolt and I observed that for a some time (couple of minutes) It receives data from other bolts while for sure those bolts is not sending him any more (because the spout was stoppled sending).
So I can say that more likely the messages are get buffered in underlying message queue (ZeroMQ,...). Is this can be true? Does this behavior ( Storm is not allocating enough CPU time to a Bolt) can be sue to the fact that I am running on a local mode??? thanks, /Shahab On Fri, Feb 28, 2014 at 1:08 PM, Enno Shioji <[email protected]> wrote: > If you literally mean due to OS not allocating CPU time to the thread when > the JVM is otherwise healthy, then no, that will be very, very, very > unlikely. > In my experience severe JVM condition can appear to have weird scheduling > issues. Did you check if the JVM is healthy? (as in has enough free heap, > doesn't have large number of threads etc.) > > > > On Fri, Feb 28, 2014 at 11:49 AM, shahab <[email protected]> wrote: > >> Hi, >> >> I am running storm on local cluster and I faced a problem. >> Is it possible that a Bolt in a Storm topology could suffer from CPU >> starvation and as result could not take the messages sent by precedent bolt >> ? >> >> /Shahab >> >> >
