[KafkaSpout] Tuples replay even if the topology is not anchored and none of the tuples fail

[Thilina Rathnayake <th...@gmail.com>]

Hi All,

I have been playing with storm and Kafka in the past couple of days and I
came
across the following problem.

I have two topologies deployed in a cluster.

1. First topology consists of a spout which emits a fixed number of
sentences and
    a `KafkaBolt` which publishes it to a topic called `sentences`.
2. Second topology consists of a `KafkaSpout` which reads from the topic
`sentences`
    and pass it to a set of bolts which split the sentence, count the words
and then report
    the counts.

I have attached the visualizations generated by Storm UI for the topologies.

First topology publishes each of it's hard coded sentences once to the topic
(I monitored this using kafka-console-consumer) and then stops publishing.

But the second topology keeps repeating the sentences received by KafkaSpout
again and again until I kill the topology. I haven't anchored any of the
bolts together
but I ack the received tuples in all the bolts.

None of the tuples fail and the bolts are not anchored. So how can I
understand
the replay of the tuples like this?

Regards,
Thilina