Storm + Kafka (pull architecture)

[Wojciech Indyk <wo...@gmail.com>]

Hello!
I was looking for information about latency of Kafka-Storm connector.
I know, that Kafka works in push-pull architecture. Therefore a client
(Storm connector here) has to ask for new data continuously (defined
time period?). Does KafkaSpout do it in any better way, like streaming
or something? How to provide data to Storm immediately? Probably it's
efficient and better choice in case of Kafka client is microbatch,
like Trident. What if I need instantaneous per-mesage processing?
Should I consider Flume instead of Kafka? Why Kafka-Storm is so
popular even if it seems to not fit together (because push-pull
architecture of Kafka)?

Kindly regards
Wojciech Indyk