SALAD Metrics Core

Implementation Overview

The library is an extension(^tm) of: spray/spray#359 When we say extension we mean an adaption of this pull request.

The main differences between this library's implementation of the Metrics implementation and that of spray/spray#359, is:

The Metrics (Timer, Counter, Meter) are sent to an Actor. It is the actor that is responsible for recording the metric into the Yammer MetricsRegistry (http://metrics.codahale.com/)

Simple usage

SBT users add the following dependency:

"com.netaporter.salad" %% "salad-metrics-core" % "0.2.5"

The pull request spray/spray#359 defines a set of spray routing aware metrics directives. These directives either record events as they come into the route (A counter), or wrap the current spray directive what could be considered kind of like an aspect. The aspect is "around" the existing directive call and at the end of the method records the event (how long the event took), i.e. a Timer.

Here is an example that times GET("/bob") and counts the number of requests to GET("/bob")

    // Get the metrics spray routing aware directive factory
val factory: MetricsDirectiveFactory = MetricsDirectiveFactory()

// Create a timer for timing GET("/bob") requests
val time = factory.timer("bobrequests").time

// Create a counter for counting GET("/bob") requests
val requestCounter = factory.counter("bobrequests").all.count

// Join the metrics together saying, I'm monitoring both the time and num of requests for GET("/bob")
val bobMetrics = time & requestCounter


// use the metrics in your routing

path("bob") {
 get {
     bobMetrics {
         complete {
             <h1>Say hello to spray</h1>
         }
     }
  }
}

Outputting the JSON metrics

The codehale metrics library includes a metrics servlet: (http://metrics.codahale.com/manual/servlets/#metricsservlet) As this metrics servlet is very java and servlet specific, a custom actor has be created that works with the MetricRegisty that the MetricsDirectiveFactory() is using.

note Note the below is in no way a demonstration of how to do spring routing. It's recommended that you put the metrics endpoint under /admin/metrics. How you do that in your spray routing, is completely upto you.

I like Tell

// get the admin actor that outputs the json
val metricsOutputActor = factory.defaultMetricsActorFactory.eventTellAdminActor()

// Set a path to for handling ("/admin/metrics") that outputs json.
//
// {"version":"3.0.0","gauges":{},"counters":{"bobrequests.successes":{"count":1}},"histograms":{},"meters":{},
// "timers":{"bobrequests":{"count":1,"max":0.054916000000000006,"mean":0.054916000000000006,"min":0.054916000000000006,
// "p50":0.054916000000000006,"p75":0.054916000000000006,"p95":0.054916000000000006,"p98":0.054916000000000006,
// "p99":0.054916000000000006,"p999":0.054916000000000006,"stddev":0.0,"m15_rate":0.0,"m1_rate":0.0,"m5_rate":0.0,
// "mean_rate":4.291145650065655,"duration_units":"seconds","rate_units":"calls/second"}}}

path("admin" / "metrics") {
  get {
      ctx => metricsOutputActor ! OutputMetrics(ctx)
  }
}

I like Ask

// get the admin actor that outputs the json
val metricsAskActor = factory.defaultMetricsActorFactory.eventAskAdminActor()

// The timeout for the ask.... this is something for you to decide.
implicit val askTimeout = Timeout(2000,TimeUnit.MILLISECONDS)

path("admin" / "metrics" / "ask") {
    get {
        onSuccess(metricsAskActor ? MetricsRequest) {
          case MetricsResponse(json : String) =>
            respondWithHeader (`Cache-Control`(`must-revalidate`, `no-store`, `no-cache`)) {
              complete(OK, json)
            }
        }
    }
}

munkyboy / salad-metrics-core