Hi LCassa, That is expected. Most modern operating systems reserve all free memory as "pagecache". Since Kafka reads/writes to the filesystem sequentially at high throughput, most of the page cache is used up at all times. That is good as well as expected.
This is different from the memory footprint of your Kafka server JVM, which can be measured by turning on some gc logging or monitoring some JVM metrics. As Jun mentions, you can also monitor the Kafka metrics he mentioned to troubleshoot issues with garbage collection pauses or large memory footprint of a Kafka server process. Thanks, Neha On Mon, Mar 24, 2014 at 10:27 PM, Cassa L <lcas...@gmail.com> wrote: > Hi, > I did not check above values for "PurgatorySize". Here is the output of > free -m when test was running. > > total used free shared buffers cached > Mem: 48356 48217 139 0 419 44584 > -/+ buffers/cache: 3212 45144 > Swap: 16383 4 16379 > > > ..LCassa > > > On Mon, Mar 24, 2014 at 8:38 PM, Jun Rao <jun...@gmail.com> wrote: > > > Are the following jmx values high on the brokers? > > > > "kafka.server":name="PurgatorySize",type="ProducerRequestPurgatory" > > "kafka.server":name="PurgatorySize",type="FetchRequestPurgatory" > > > > Thanks, > > > > Jun > > > > On Mon, Mar 24, 2014 at 12:38 PM, Cassa L <lcas...@gmail.com> wrote: > > > > > Hi, > > > We have been doing some evaluation testing against Kafka.We have 48GB > > RAM > > > on each broker. I created 3 broker cluster and one zookeeper and sent > > > 10,000Messages/seconds to this cluster, continuously. Payload is very > > > small, less than Kilobyte. Consumer was readind 5,000Messages per > second. > > > > > > What I noticed that, after couple of hours, entire memory was being > > used. > > > There was no free memory available on those machines. I believe Kafka > > does > > > use Off-Heap memory, how do reduce this consumption? How can I avoid > > > OutOfMemory problem if this happens ? > > > Are there any setting I need to do? I was playing with some > configuration > > > parameters and here are they - > > > log.flush.interval.messages=100000 (default > > > 10000) > > > log.flush.interval.ms=60000 > > > (default 1000) > > > log.default.flush.scheduler.interval.ms=10000 (default > > 3000) > > > log.cleanup.interval.mins=120 > > (default > > > 10) > > > > > > Thanks, > > > LCassa > > > > > >