Hi Claudio, The configuration of task.opts is unfortunately a known annoyance. Plans to improve the situation are here: https://issues.apache.org/jira/browse/SAMZA-109 -- see a comment on that issue for a workaround that you can use in the meantime.
Best, Martin On 11 May 2014, at 12:52, Claudio Martins <[email protected]> wrote: > Hi all, > > we have a task that it needs to consume lots of message due to the size of > the messages and all the Json parsing. > > Running with the default of Xmx768 seems very small and YARN has been > killing the task and restarting constantly. > > Playing around with the task.opts parameter, I can setup a new maximum > value, however, when the container is started all the other parameters seem > to get lost and the log files get a little messy (stderr now has the > content of samza.container.log, for example). > > This is the default without setting up the task.opts: > > -Xmx768M -XX:+PrintGCDateStamps > -Xloggc:/usr/hadoop/logs/userlogs/application_1399307125726_0035/container_1399307125726_0035_01_000034/gc.log > -Dsamza.log.dir=/usr/hadoop/logs/userlogs/application_1399307125726_0035/container_1399307125726_0035_01_000034 > > > I've setup the task.opts like this: > > -Xmx1024M -XX:+PrintGCDateStamps -Xloggc:logs/gc.log > > And then the command line becomes only: > > -Xmx1280M -XX:+PrintGCDateStamps -Xloggc:logs/gc.log -cp (everything else > here) > > > Any tip on how to setup this property correctly? > > > > > - Claudio Martins > linkedin: www.linkedin.com/in/martinsclaudio
