Fiber Failure Metric Double Counts Failures

[khajavi]

I expect this example to print 5 for String causes and 2 for null pointer exception causes.

import zio._

object RuntimeMetricsExample extends ZIOAppDefault {
  override val bootstrap =
    Runtime.enableRuntimeMetrics

  val nullErrors   =
    ZIO.foreachParDiscard(1 to 2)(_ => ZIO.attempt(null.toString))

  val customErrors =
    ZIO.foreachParDiscard(1 to 5)(_ => ZIO.fail("Custom application error"))

  def run = {
    nullErrors <&> customErrors
  }.onError { _ =>
    Metric.runtime.fiberFailureCauses.value
      .flatMap(failureCauses =>
        ZIO
          .foreachDiscard(failureCauses.occurrences) { case (cause, count) =>
            Console.printLine(s"Failure cause: $cause, Count: $count")
          }
          .orDie
      )
  }.ignore
}

But the output is:

Failure cause: java.lang.String, Count: 10
Failure cause: java.lang.NullPointerException, Count: 4

[kyri-petrou]

The issue here is that when we join the results of the two fibers created by <&> we're essentially creating a new Cause that concats all the failures and we count them again.

One way to fix this would be to filter on the fiberId in the Cause's traces, but that would only work with ZIO.fail / ZIO.attempt etc, but not with Exit.fail. In the case of Exit.fail we would have to either double-count or not include it in metrics at all.