Metrics

Meter is an entry point to the metrics capabilities and instrumentation.

How to get the Meter

Currently, otel4s has a backend built on top of OpenTelemetry Java. Add the following configuration to the favorite build tool:

Add settings to the build.sbt:

libraryDependencies ++= Seq(
  "org.typelevel" %% "otel4s-oteljava" % "0.11.1", // <1>
  "io.opentelemetry" % "opentelemetry-exporter-otlp" % "1.44.1" % Runtime, // <2>
  "io.opentelemetry" % "opentelemetry-sdk-extension-autoconfigure" % "1.44.1" % Runtime // <3>
)
javaOptions += "-Dotel.java.global-autoconfigure.enabled=true" // <4>

Add directives to the *.scala file:

//> using dep "org.typelevel::otel4s-oteljava:0.11.1" // <1>
//> using dep "io.opentelemetry:opentelemetry-exporter-otlp:1.44.1" // <2>
//> using dep "io.opentelemetry:opentelemetry-sdk-extension-autoconfigure:1.44.1" // <3>
//> using javaOpt "-Dotel.java.global-autoconfigure.enabled=true" // <4>
  1. Add the otel4s-oteljava library
  2. Add an OpenTelemetry exporter. Without the exporter, the application will crash
  3. Add an OpenTelemetry autoconfigure extension
  4. Enable OpenTelemetry SDK autoconfigure mode

Once the build configuration is up-to-date, the Meter can be created:

import cats.effect.IO
import org.typelevel.otel4s.metrics.Meter
import org.typelevel.otel4s.oteljava.OtelJava

OtelJava.autoConfigured[IO]().evalMap { otel4s =>
  otel4s.meterProvider.get("com.service").flatMap { implicit meter: Meter[IO] =>
    val _ = meter // use meter here
    ???
  }
}

Available instruments

The instruments are split into two categories: synchronous and asynchronous (observable).

The terms synchronous and asynchronous have nothing to do with asynchronous programming. The naming follows the OpenTelemetry specification.

To create an instrument, you must specify the measurement type. The Long and Double are available out of the box.

import cats.effect.IO
import org.typelevel.otel4s.metrics.{Counter, Meter}

@annotation.nowarn
val meter: Meter[IO] = ???

val doubleCounter: IO[Counter[IO, Double]] =
  meter.counter[Double]("double-counter").create

val longCounter: IO[Counter[IO, Long]] =
  meter.counter[Long]("long-counter").create

The recommended measurement types per instrument:

Instrument Type Measurement type
Counter Synchronous Long
UpDownCounter Synchronous Long
Histogram Synchronous Double
ObservableCounter Asynchronous Long
ObservableGauge Asynchronous Double
ObservableUpDownCounter Asynchronous Long

Synchronous instruments

Synchronous instruments are meant to be invoked inline with application/business processing logic. For instance, an HTTP client might utilize a counter to record the number of received bytes.

The synchronous instruments are:

The following example tracks the number of users missing in the storage and the duration of the retrieval:

import java.util.concurrent.TimeUnit

import cats.Monad
import cats.effect.{Concurrent, MonadCancelThrow, Ref}
import cats.syntax.applicative._
import cats.syntax.flatMap._
import cats.syntax.functor._
import org.typelevel.otel4s.metrics.{Counter, Histogram, Meter}

case class User(email: String)

class UserRepository[F[_]: MonadCancelThrow](
    storage: Ref[F, Map[Long, User]],
    missingCounter: Counter[F, Long],
    searchDuration: Histogram[F, Double]
) {

  def findUser(userId: Long): F[Option[User]] =
    searchDuration.recordDuration(TimeUnit.SECONDS).surround(
      for {
        current <- storage.get
        user    <- Monad[F].pure(current.get(userId))
        _       <- missingCounter.inc().whenA(user.isEmpty)
      } yield user
    )

}

object UserRepository {

  def create[F[_]: Concurrent: Meter]: F[UserRepository[F]] = {
    for {
      storage  <- Concurrent[F].ref(Map.empty[Long, User])
      missing  <- Meter[F].counter[Long]("user.search.missing").create
      duration <- Meter[F].histogram[Double]("user.search.duration").withUnit("s").create
    } yield new UserRepository(storage, missing, duration)
  }

}

Asynchronous (observable) instruments

Asynchronous instruments offer users the ability to register callback functions, which are only triggered on demand. For example, an asynchronous gauge can be used to collect the temperature from a sensor every 15 seconds, which means the callback function will only be invoked every 15 seconds.

The asynchronous instruments are:

The following example shows how to collect MBean metrics:

import java.lang.management.ManagementFactory
import javax.management.ObjectName

import cats.effect.{Resource, Sync}
import org.typelevel.otel4s.metrics.Meter

object CatsEffectMetrics {

  private val mbeanName = new ObjectName(
    "cats.effect.metrics:type=CpuStarvation"
  )

  def register[F[_]: Sync: Meter]: Resource[F, Unit] =
    for {
      mBeanServer <- Resource.eval(
        Sync[F].delay(ManagementFactory.getPlatformMBeanServer)
      )
      _ <- Meter[F]
        .observableCounter[Long]("cats_effect.runtime.cpu_starvation.count")
        .createWithCallback { cb =>
          cb.record(
            mBeanServer
              .getAttribute(mbeanName, "CpuStarvationCount")
              .asInstanceOf[Long]
          )
        }
    } yield ()

}