Http4s Finagle/Why Finagle x Http4s?
Why Finagle x Http4s?

Why Finagle x Http4s?

Finagle is an extensible RPC system that Twitter running on, there are bunch of production ready things to use, without needing particular setup, batteries included such as:

Zipkin distributed tracing

So if all your microservices are running on Finagle, it is pretty much 0 config to get Zipkin just works. you can query a request's journey through all your microservices by just query it's trace id, interesting data will be summarized for you, such as the percentage of time spent in which service, and whether or not operations failed.

Metrics

Finagle will automatically generate metrics for server and client when you enable them with

.withHttpStats


   
 
   
There are some adapter ready to use to expose to different format, such as Prometheus
 
   
Circuit Breaker x Load Balancer x Retry
 
    
   
Since Finagle is RPC system, it is very easy to detect error accurately, you can also apply custom classifier to identity different kinds of error
 
   
Twitter Server
 
   
If we run Finagle Service on a Twitter Server, then we will get
 
   
     
    
  • Logging with dynamo log level control
    •  
    
  • flag and feature toggle
    •  
    
  • a Admin portal that can monitor metrics and histograms
    •  
   
 
   
Finagle x Http4s
 
   
The pain point is when there are lot of existing Finagle RPC services running on production already, and when I introduced Http4s service, I need to spend too many time to make Zipkin, metrics, circuit breaker work the same way with the existing services, but if I give up new service will be couple with Finagle/Finatra dsl forever.
 
   
Now with http4s-finagle, we can actually take advantage of both, create business with Http4s, and run them on Finagle to keep Devops practice consistent.