devbv commented on issue #3072: Enhance kafka message provider URL: https://github.com/apache/incubator-openwhisk/pull/3072#issuecomment-349975958 @markusthoemmes @jeremiaswerner I can't write test case for almost scenarios because they are intermittent issues from Kafka fail-over and rebalancing and it is so hard to recreate it manually. My purpose was trying to success all user request by handling exceptions and adding fallbacks, even there are node fails in kafka cluster. Well tested client like akka-stream also suffers same issues but it just propagates erros to user code. If it is ok to allow all errors and response user request as a internal error when kafka issues occur, I definitely agree with that we don't need to handle exceptions and it's quite large operation. So if you thought in that way, I would get rid of all error handling and retry logics! And in that case, our kafka shooting up test should be changed to something like "Allow at most 3 errors for 10 request". So give me opinons, I'll gladly follow yours. Meanwhile, Poll() should be handled by wakeup() and WakeupException. Because it could ignore timeout and hangs until one of failed nodes come back again. (https://github.com/akka/reactive-kafka/blob/4034e436ab387fcd00077e6fc98da56713bfe28a/core/src/main/scala/akka/kafka/KafkaConsumerActor.scala#L275).
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
