Spark streaming uses micro-batching to achieve near real-time processing. If latency must be under 1 sec, you should consider Storm Core. There are many factors you should consider, e.g. latency, data processing requirement (machine learning, SQL access, graph processing etc.), YARN support, other language compatibility and maturity. The following links may help.
http://statrgy.com/2015/05/14/storm-vs-spark/
http://realtime-cachedmind.tumblr.com/post/89974796387/real-time-processing-storm-trident-vs
http://ingest.tips/2015/06/24/real-time-analytics-with-kafka-and-spark-streaming/
|