Hmm but when new people try Kafka they will probably expect near zero latency and this will give them 1s.
On Oct 20, 2011, at 8:39 AM, Jay Kreps <jay.kr...@gmail.com> wrote: > Filed https://issues.apache.org/jira/browse/KAFKA-164 for this. > > -Jay > > On Thu, Oct 20, 2011 at 7:56 AM, Jay Kreps <jay.kr...@gmail.com> wrote: > >> This is a good point. I recommend we change the default to something higher >> and set time-based flush to 1 sec. >> >> -Jay >> >> >> On Thu, Oct 20, 2011 at 7:09 AM, Tim Lossen <t...@lossen.de> wrote: >> >>> ok, never mind, we found the culprit. the default server config >>> contains the following setting: >>> >>>> #the number of messages to accept without flushing the log to disk >>>> log.flush.interval=1 >>> >>> we changed this value to 10000 now, and the performance is excellent. >>> maybe not the most sensible default value? >>> >>> cheers >>> tim >>> >>> >>> On 2011-10-20, at 11:00 , Tim Lossen wrote: >>> >>>> hello, >>>> >>>> we have been playing around with kafka the last few days, and on >>>> our developer machines (osx 10.6) everythings looks fine. but on >>>> linux (ubuntu 11.04) we get very strange behaviour: >>>> >>>> - the producer seems to get stuck after a few thousand messages >>>> - the consumer seems to run very slowly >>>> >>>> (kafka 0.6, java 1.6.0, default server + zookeeper configuration, >>>> simple producer and consumer scripts written in ruby -- everything >>>> running on same machine.) >>>> >>>> maybe there is something wrong with the os-level network config? >>>> other ideas? any hints / suggestions welcome. >>>> >>>> cheers >>>> tim >>>> >>>> -- >>>> http://tim.lossen.de >>>> >>>> >>>> >>> >>> -- >>> http://tim.lossen.de >>> >>> >>> >>> >>
