Details
-
Bug
-
Status: Closed
-
Major
-
Resolution: Fixed
-
0.8.0
-
None
-
None
Description
Hey Guys,
It seems that, by default, producers in 0.8 are async, and have a default queue.enqueue.timeout.ms of 0. This means that anyone who reads messages faster than they're producing them will likely end up eventually hitting this exception:
Exception in thread "Thread-3" kafka.common.QueueFullException: Event queue is full of unsent messages, could not send event: KeyedMessage(PageViewEventByGroupJson,Missing Page Group,java.nio.HeapByteBuffer[pos=0 lim=125 cap=125])
at kafka.producer.Producer$$anonfun$asyncSend$1.apply(Producer.scala:111)
at kafka.producer.Producer$$anonfun$asyncSend$1.apply(Producer.scala:89)
at scala.collection.IndexedSeqOptimized$class.foreach(IndexedSeqOptimized.scala:34)
at scala.collection.mutable.WrappedArray.foreach(WrappedArray.scala:32)
at kafka.producer.Producer.asyncSend(Producer.scala:89)
at kafka.producer.Producer.send(Producer.scala:77)
As it says in https://cwiki.apache.org/KAFKA/kafka-mirroring.html, this can result in losing messages, and nasty exceptions in the logs. I think the better default is setting queue.enqueue.timeout.ms to -1, which will just block until the queue frees up.