Reactive Principles

By Barış Daylık
http://kod.io

Transcript

1. Reactive Principles Barış Can Daylık baris@vngrs.com
2. Agenda • Current application requirements • Traditional Solutions • New Architecture (Reactive) – Event-Driven – Scalable – Resilient – Responsive • Final Words
3. Current application requirements • Before – Tens of servers – Seconds of response time – Hours of offline maintenance – Gigabytes of data • Now – Diverse platforms (i.e. from mobile devices to cloud-based clusters on thousands of multicore processors) – Millisecond to microsecond response times – 100% uptime – Petabytes of data
4. Traditional Solutions • Dedicated Bare-Metal Servers • Vertical Scaling (buying larger servers) • Multi-threading for concurrency • Adding servers are complex and expensive • Not easy to manage
5. Traditional Solutions • Traditional Server Architecture – Multi Threading – Shared Mutable State • Synchronization/Locks • Non-Deterministic/Race Conditions • Code is hard to understand and maintain – Blocking a single thread • Resource consuming • High wake-up cost
6. New Architecture • Satisfies todays demands • Reactive Applications – Event-driven – Scalable – Resilient (Fault-Tolerant) – Responsive
7. New Architecture • Satisfies todays demands • Reactive Applications – Event-driven – Scalable – Resilient (Fault-Tolerant) – Responsive
8. New Architecture • Reactive applications – react to events (Event-Driven) – react to load (Scalable) – react to failure (Resilient) – react to users (Responsive)
9. Event-Driven • Interacts through producing and consuming events • Asynchronous message passing – Easy parallelization • Loose Coupling – Separated Producer and Consumer – Hiding the details of event propagation and communication – Location Independence – Easy to extend – Easy to maintain – Providing flexibility while programming
10. Event-Driven • Non-Blocking – Sharing the same thread among multiple consumers – Higher throughput – Lower latency – Effective hardware utilization and lower costs • Push based – Resources not wasted by polling or waiting on data
11. Event-Driven • Application must be event- driven from UI down to the Database.
12. Event-Driven • Even if %5 of the whole application is sequential, total speedup will be limited to 20x, according to Amdahl's law. Amdahl's Law
13. Scalable • Scalable application – Capable of being easily expanded or upgraded on demand based on its usage – Scale up or down without redesign/rewrite – Cost effective on cloud deployments as it can grow or shrink elastically based on usage
14. Scalable • Loose coupling and location independence of event- driven system makes it easy to scale • Easy to add more event consumers • No difference between vertical/horizontal scaling
15. Resilient • Recover quickly from failures • Application downtime – Nightmare of a business – Halts revenue stream – Causes unhappy customers and poor reputation
16. Resilient • Traditional Applications – Ignored fail scenarios – Exception handling • Not all problems can be dealt with where they occur – Clustered servers • Failover for whole application • Recovery is time consuming thus costly • Potential risk of cascading failures
17. Resilient • Reactive Applications – Resilience is part of the design – Failure is a first class construct • Easier to react and manage • Auto healing of component instead of application failover
18. Resilient • To manage failure – Isolate to prevent cascading failures – Observe to detect it whenever it happens, not when its too late • Event-driven model to the rescue – Full isolation as a result of loose coupling – Failures can be packaged and sent as events where other components decides how to deal with it – Helps keeping business flow clear and lets you deal with failure in a more rigid and planned way – Location transparent
19. Responsive • Its not HTML5 responsive design :) • Responsive Applications are – Quick to respond and react – Responds fast even in case of failure • Not responding in a timely manner – Might cause total failure (Medical Systems) – Decreased utility (Finance, Telco, Internet)
20. Responsive • Observable Models – Triggers events on state changes – Makes it possible to synchronize between remote clients • Event Streams – Asynchronous and non-blocking communications
21. Final Words • Reactive applications has four traits – Event-Driven – Scalable – Resilient – Responsive • Complements each other forming a whole • Addresses common problems with current challenges in application design
22. Don't forget to check these out • Reactive Manifesto : http://www.reactivemanifesto.org/ • ReactiveMongo : http://reactivemongo.org/ • Akka : http://akka.io/ • Play : http://www.playframework.com/ • Reactive Activator Template http://typesafe.com/activator/template/tcp-async
23. We Are Hiring! info@vngrs.com