Hi Adrian, Do you know why the tuples are timing out. Do you see any
failures in downstream bolts. I think fixing the timeouts is a better
approach than ignoring the failed tuples. It looks to be your downstream
bolts unable to ack a tuple in time may be increasing the parallelism of
those bolts might help and also check your JVM size for workers if the
"-Xmx" value is right for your workload. -Harsha


On Tue, Nov 25, 2014, at 06:42 PM, 이승진 wrote:
> Hi all,


>


> currently I'm using kafkaspout 0.9.2 as a topology spout(not trident).


>


> and all the execute methods are wrapped with try/catch.


>


> but sometimes tuple times out, and I think they are regareded as
> faild tuples


>


> so the counter of kafka spout failed tuple increases, and at some
> point, the spout itself hangs.


>


> when I look up the logs, only relevant lines are like below


> s.k.PartitionManager [INFO] Skipping failed tuple at offset=2617297034
>     because it's more than maxOffsetBehind=100000 behind
>     _emittedToOffset=2618171999


>


> I think some of them are not skipped and stay in queue which cause
> spout to be blocked


>


> do I have to set SpoutConfig.maxOffsetBehind=0 to ignore time out
> failures? or any other suggestion?


>


> currently, tuple time out is set to 60 secs


>


> thanks in advance
>
> 안녕하세요 분산시스템개발랩 이승진입니다.
>
>
>
> 감사합니다.
>
> 이승진 드림

> 이 메일은 나눔글꼴로 작성되었습니다. 설치하기[1]
>
>
> *이승진* Adrian Seungjin Lee Distributed System Development Lab /
> Researcher
>
> NAVER Green Factory, 6, Buljeong-ro, Bundang-gu, Seongnam-si,
> Gyeonggi-do, 463-867, Korea *Tel* 031-784-1274 *Mobile* 010-7141-7134
> *Email* [email protected] *Messenger*
>
> NHN
>



Links:

  1. http://hangeul.naver.com

Reply via email to