Tracing

Tracer is an entry point to the tracing capabilities and instrumentation. It provides various functionalities for creating and managing spans, extracting context from carriers, propagating context downstream, and more.

How to get the Tracer

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

libraryDependencies ++= Seq(
  "org.typelevel" %% "otel4s-oteljava" % "0.12.0", // <1>
  "io.opentelemetry" % "opentelemetry-exporter-otlp" % "1.49.0" % Runtime, // <2>
  "io.opentelemetry" % "opentelemetry-sdk-extension-autoconfigure" % "1.49.0" % Runtime // <3>
)
javaOptions += "-Dotel.java.global-autoconfigure.enabled=true" // <4>

//> using dep "org.typelevel::otel4s-oteljava:0.12.0" // <1>
//> using dep "io.opentelemetry:opentelemetry-exporter-otlp:1.49.0" // <2>
//> using dep "io.opentelemetry:opentelemetry-sdk-extension-autoconfigure:1.49.0" // <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 Tracer can be created:

import cats.effect.IO
import org.typelevel.otel4s.trace.Tracer
import org.typelevel.otel4s.oteljava.OtelJava

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

Creating a span

You can use the span or spanBuilder API to create a new span.

The tracer automatically determines whether to create a child span or a root span based on the presence of a valid parent in the tracing context. If a valid parent is available, the new span becomes a child of it. Otherwise, it becomes a root span.

Here's how you can do it:

import cats.Monad
import cats.effect.Ref
import cats.syntax.flatMap._
import cats.syntax.functor._
import org.typelevel.otel4s.Attribute
import org.typelevel.otel4s.trace.Tracer

case class User(email: String)

class UserRepository[F[_]: Monad: Tracer](storage: Ref[F, Map[Long, User]]) {

  def findUser(userId: Long): F[Option[User]] =
    Tracer[F].span("find-user", Attribute("user_id", userId)).use { span =>
      for {
        current <- storage.get
        user <- Monad[F].pure(current.get(userId))
        _ <- span.addAttribute(Attribute("user_exists", user.isDefined))
      } yield user
    }

}

Starting a root span

A root span is a span that is not a child of any other span. You can use Tracer[F].rootScope to wrap an existing effect or Tracer[F].rootSpan to explicitly start a new root span:

import cats.Monad
import cats.syntax.flatMap._
import cats.syntax.functor._

class UserRequestHandler[F[_]: Tracer: Monad](repo: UserRepository[F]) {
  private val SystemUserId = -1L

  def handleUser(userId: Long): F[Unit] =
    Tracer[F].rootScope(activateUser(userId))

  def handleUserInternal(userId: Long): F[Unit] =
    Tracer[F].rootSpan("handle-user").surround(activateUser(userId))

  private def activateUser(userId: Long): F[Unit] =
    for {
      systemUser <- repo.findUser(SystemUserId)
      user <- repo.findUser(userId)
      _ <- activate(systemUser, user)
    } yield ()

  private def activate(systemUser: Option[User], target: Option[User]): F[Unit] = {
    val _ = (systemUser, target) // some processing logic
    Monad[F].unit
  }
}

While the behavior seems similar, the outcome is notably different:

1. Tracer[F].rootScope(activateUser(userId)) will create two independent root spans:

gantt
    dateFormat HH:mm:ss
    axisFormat %H:%M:%S

    section Span 1
    find-user { user_id = -1 }     :done, a1, 00:00:00, 00:00:10
    section Span 2
    find-user { user_id = 123 }    :done, a2, 00:00:00, 00:00:10

1. Tracer[F].rootSpan("handle-user").surround(activateUser(userId)) will create two child spans:

gantt
    dateFormat HH:mm:ss
    axisFormat %H:%M:%S

    section Spans
    handle-user                    :done, a1, 00:00:00, 00:00:10
    find-user { user_id = -1 }     :done, a3, 00:00:01, 00:00:10
    find-user { user_id = 123 }    :done, a2, 00:00:01, 00:00:10

Running effect without tracing

If you want to disable tracing for a specific section of the effect, you can use the Tracer[F].noopScope. This creates a no-op scope where tracing operations have no effect:

class InternalUserService[F[_]: Tracer](repo: UserRepository[F]) {

  def findUserInternal(userId: Long): F[Option[User]] =
    Tracer[F].noopScope(repo.findUser(userId))

}

Starting an unmanaged span

The Tracer[F].span(...) automatically manages the lifecycle of the span. Tracer[F].span("...").startUnmanaged creates a span that must be ended manually by invoking end. This strategy can be used when it's necessary to end a span outside the scope (e.g. async callback).

A few limitations:

1. An unfinished span remains active indefinitely

In the following example, the unmanaged span has never been terminated:

import org.typelevel.otel4s.trace.StatusCode

def leaked[F[_]: Monad: Tracer]: F[Unit] =
  Tracer[F].spanBuilder("manual-span").build.startUnmanaged.flatMap { span =>
    span.setStatus(StatusCode.Ok, "all good")
  }

Properly ended span:

def ok[F[_]: Monad: Tracer]: F[Unit] =
  Tracer[F].spanBuilder("manual-span").build.startUnmanaged.flatMap { span =>
    span.setStatus(StatusCode.Ok, "all good") >> span.end
  }

2. The span isn't propagated automatically

Consider the following example:

def nonPropagated[F[_]: Monad: Tracer]: F[Unit] = 
  Tracer[F].span("auto").surround {
    // 'unmanaged' is the child of the 'auto' span
    Tracer[F].span("unmanaged").startUnmanaged.flatMap { unmanaged =>
      // 'child-1' is the child of the 'auto', not 'unmanaged'  
      Tracer[F].span("child-1").use_ >> unmanaged.end
    }
  }

The structure is:

gantt
    dateFormat HH:mm:ss
    axisFormat %H:%M:%S

    section Spans
    auto (root)                    :done, a1, 00:00:00, 00:00:10
    unmanaged (child of 'auto')    :done, a2, 00:00:02, 00:00:09
    child-1 (child of 'auto')      :done, a3, 00:00:02, 00:00:08

Use Tracer[F].childScope to create a child of the unmanaged span:

def propagated[F[_]: Monad: Tracer]: F[Unit] = 
  Tracer[F].span("auto").surround {
    // 'unmanaged' is the child of the 'auto' span
    Tracer[F].span("unmanaged").startUnmanaged.flatMap { unmanaged => 
      Tracer[F].childScope(unmanaged.context) {
        // 'child-1' is the child of the 'unmanaged' span
        Tracer[F].span("child-1").use_ >> unmanaged.end
      }
    }
  }

The structure is:

gantt
    dateFormat HH:mm:ss
    axisFormat %H:%M:%S

    section Spans
    auto (root)                         :done, a1, 00:00:00, 00:00:10
    unmanaged (child of 'auto')         :done, a2, 00:00:02, 00:00:09
    child-1 (child of 'unmanaged')      :done, a3, 00:00:03, 00:00:08

Tracing a resource

You can use Tracer[F].span("resource").resource to create a managed span.

Consider the following example:

import cats.effect._
import cats.syntax.functor._
import org.typelevel.otel4s.trace.{Tracer, SpanOps}

def withResource[F[_]: Async: Tracer]: F[Unit] =
  Tracer[F].span("my-resource-span").resource.use { case SpanOps.Res(_, _) =>
    Tracer[F].currentSpanContext.void // returns `None`
  }

you must evaluate the inner effect within the trace to propagate span details:

def withResource[F[_]: Async: Tracer]: F[Unit] =
  Tracer[F].span("my-resource-span").resource.use { case SpanOps.Res(_, trace) =>
    trace(Tracer[F].currentSpanContext).void // returns `Some(SpanContext{traceId="...", })`
  }

Structured spans

You can achieve structured spans in the following way:

class Connection[F[_]: Tracer] {
  def use[A](f: Connection[F] => F[A]): F[A] =
    Tracer[F].span("use_conn").surround(f(this))
}

object Connection {
  def create[F[_]: Async: Tracer]: Resource[F, Connection[F]] = 
    Resource.make(
      Tracer[F].span("acquire").surround(Async[F].pure(new Connection[F]))
    )(_ => Tracer[F].span("release").surround(Async[F].unit))
}

class App[F[_]: Async: Tracer] {
  def withConnection[A](f: Connection[F] => F[A]): F[A] =
    (for {
      r <- Tracer[F].span("resource").resource
      c <- Connection.create[F].mapK(r.trace)
    } yield (r, c)).use { case (res, connection) =>
      res.trace(Tracer[F].span("use").surround(connection.use(f)))
    }  
}

The spans structure is:

gantt
    dateFormat HH:mm:ss
    axisFormat %H:%M:%S

    section Spans
    resource        :done, a1, 00:00:00, 00:00:10
    acquire         :done, a2, 00:00:00, 00:00:03
    use             :done, a3, 00:00:03, 00:00:08
    use_conn        :done, a4, 00:00:04, 00:00:08
    release         :done, a5, 00:00:08, 00:00:10

Acquire and release spans

You can also use res.trace in combination with Resource#mapK to trace the acquire and release steps of a resource:

class Transactor[F[_]]
class Redis[F[_]]

def createTransactor[F[_]: Async: Tracer]: Resource[F, Transactor[F]] =
  Resource.make(
    Tracer[F].span("transactor#acquire").surround(Async[F].pure(new Transactor[F]))
  )(_ => Tracer[F].span("transactor#release").surround(Async[F].unit))
  
def createRedis[F[_]: Async: Tracer]: Resource[F, Redis[F]] =
  Resource.make(
    Tracer[F].span("redis#acquire").surround(Async[F].pure(new Redis[F]))
  )(_ => Tracer[F].span("redis#release").surround(Async[F].unit))
  
def components[F[_]: Async: Tracer]: Resource[F, (Transactor[F], Redis[F])] =
  for {
    r <- Tracer[F].span("app_lifecycle").resource
    tx <- createTransactor[F].mapK(r.trace)
    redis <- createRedis[F].mapK(r.trace)
  } yield (tx, redis)

def run[F[_]: Async: Tracer]: F[Unit] =
  components[F].use { case (_ /*transactor*/, _ /*redis*/) =>
    Tracer[F].span("app_run").surround(Async[F].unit)
  }

The spans structure is:

gantt
    dateFormat HH:mm:ss
    axisFormat %H:%M:%S

    section app_lifecycle
    app_lifecycle      :done, a1, 00:00:00, 00:00:26
    transactor#acquire :done, a2, 00:00:00, 00:00:06
    redis#acquire      :done, a3, 00:00:06, 00:00:10
    redis#release      :done, a4, 00:00:15, 00:00:19
    transactor#release :done, a5, 00:00:19, 00:00:26
    
    section app_run
    app_run           :done, a6, 00:00:10, 00:00:15

Both app_run and app_lifecycle are unique and not linked to each other.