tips to design your stream processing setup.
what all can go wrong, how to instrument it.

Outline

introduction to a production setup that handles billions of events per week through our home-grown apache kafka client, process the stream with storm - and then aggregate into postgres. will also share pro’s and con’s of building our own kafka client as opposed to re-using one.

Speaker bio

Bosky (@bhaskerkode) leads a product engg team at Helpshift & works on erlang, clojure and golang.

building distributed systems since ‘06 across edtech, adtech & mobile in erlang, clojure & go.
=> http://in.linkedin.com/in/bhaskerkode & http://slideshare.net/bosky101

Built a kafka producer/micro-service used in production at helpshift, layer, and several other companies.
=> http://github.com/helpshift/ekaf
( Recommended by Apache kafka https://cwiki.apache.org/confluence/display/KAFKA/Clients )

Using storm in production for sentiment analysis, topic extraction, naive bayes classification, etc.
Eager to learn more about best practices in storm deployment/management. Incidentally another part that feeds into this system is in golang, uses the Shopify kafka producer.

Links

• http://github.com/helpshift/ekaf
• https://coderwall.com/p/1lyfxg/parsing-the-kafka-protocol-with-erlang-pattern-matching-ftw

Slides

https://www.dropbox.com/s/c4rwq1tiu7nxvpd/fifthel-15-kafka-storm-at-helpshift-bhaskerkode.png?dl=0