Paul, That behavior is currently expected, see https://issues.apache.org/jira/browse/KAFKA-1788. There are currently no client-side timeouts in the new producer, so the message just sits there forever waiting for the server to come back so it can try to send it.
If you already have tests for a variety of failure modes it might be helpful to contribute them to the WIP patch in that JIRA. Currently the patch only adds one unit test for RecordAccumulator, a few tests using the full server would be an improvement. -Ewen On Tue, Jan 6, 2015 at 7:54 AM, Paul Pearcy <ppea...@gmail.com> wrote: > Hello, > I have some of my own test cases very similar to the ones > in ProducerFailureHandlingTest. > > I moved to the new producer against 0.8.1.1 and all of my test cases around > disconnects fail. Moving to use 0.8.2-beta on server side things succeed. > > Here is an example test: > - Healthy producer/server setup > - Stop the server > - Send a message > - Call get on the future and it never returns. Doesn't matter if the server > is started again or remains stopped > > So, I believe that while 0.8.2 producer is compatible with 0.8.1.1 in the > happy case, but fails to return futures in cases where disconnects have > occurred. > > Is it possible to run the test cases in ProducerFailureHandlingTest.scala > against 0.8.1.1? > > Thanks, > Paul > -- Thanks, Ewen
