I was wondering what the first step would be to integrating this into a kubernetes cluster. Does anyone know of a tutorial that applies this library or something similar?

Below is a simple reproduction of the issue, tested with version 0.23.12

Reproduction

import cats._
import cats.implicits._
import cats.effect._
import cats.effect.unsafe.implicits._
import io.prometheus.client._
import scala.jdk.CollectionConverters._
import org.http4s.metrics.prometheus._

def printCR(name: String, cr: CollectorRegistry): IO[Unit] = 
  IO.println(s"$name: ${cr.metricFamilySamples.asScala.map(_.name).take(3).mkString("\n  ")}")

def go(name: String, cr: CollectorRegistry): IO[Unit] = {
  IO.println(name) *>
    printCR(s"$name Start", cr) *>
    PrometheusExportService.addDefaults[IO](cr)
      .use { _ => 
        printCR(s"$name After", cr)
      }
  }

{
  go("New Collector", new CollectorRegistry()) *>
  go("Default Collector", CollectorRegistry.defaultRegistry) 
}.unsafeRunSync()

Yields

New Collector
New Collector Start: 
New Collector After: jvm
  process_cpu_seconds
  process_start_time_seconds
Default Collector
Default Collector Start: 
java.lang.IllegalArgumentException: Failed to register Collector of type ClassLoadingExports: The Collector exposes the same name multiple times: jvm_classes_loaded
  at io.prometheus.client.CollectorRegistry.assertNoDuplicateNames(CollectorRegistry.java:71)
  at io.prometheus.client.CollectorRegistry.register(CollectorRegistry.java:51)
  at io.prometheus.client.Collector.register(Collector.java:307)
  at org.http4s.metrics.prometheus.Prometheus$.$anonfun$registerCollector$1(Prometheus.scala:287)
  at println @ printCR(<console>:9)
  at make @ org.http4s.metrics.prometheus.Prometheus$.registerCollector(Prometheus.scala:287)
  at make @ org.http4s.metrics.prometheus.Prometheus$.registerCollector(Prometheus.scala:287)
  ... 5 elided

Hi team,

I discovered a weird behaviour if metrics are enabled on an HttpRoutes value that is composed with some other value.

For example let's say there are apiRoutes and adminRoutes defined and compose like this:

val metrics: MetricsOps[F] = ???
val apiRoutes: HttpRoutes[F] = ???
val adminRoutes: HttpRoutes[F] = ???

val apiRoutesWithMetrics = Metrics[IO](metrics)(apiRoutes)

httpApp = (
        apiRoutesWithMetrics <+> adminRoutes.  //  <+> SemigroupK Cats
      ).mapF(_.getOrElseF(NotFound()))

return httpApp // <- this is used to create the http4s webserver

Whenever there is a request meant for the adminRoutes the metrics defined for apiRoutesWithMetrics gets incremented for 4xx requests although the request is send back to the client with 200 OK status.

If the routes are composed the other way around adminRoutes <+> apiRoutesWithMetrics this behaviour does not happen.

I'm guessing is because the request goes through the first routes and exists with NotFound (400) and the metrics are incremented and then the request go to the admin routes and there is a route defined but the metrics are already incremented.

I'm not saying it's a bug in the library but maybe there can be added to the documentation.
I'm happy to add it.

What is your opinion on this?

Regards,
Robert

Currently the process of enumerating the metrics from all registered collectors is wrapped in Sync[F].delay:

If this is a lengthy / IO-bound operation, this might starve the compute pool of the server. It seems a better choice is to use Sync[F].blocking for that task.

Follow-up of discussion on Discord: https://discord.com/channels/632277896739946517/632286375311573032/1164491603017207808

Hey, for someone with only limited knowledge around all this Resource/F things it would be extremely helpful to get a simple example in the Readme.md on how to actually use it, i.e. how to combine the PrometheusExportService with an existing http4s App…

I created metered client.
The goal was to calculate sum(increase(_requests_total)
Result is that the first request to the client is not counted, because counters are not available at zero

It would make getting started easier if this repo has a grafana dashboard that demonstrates how all the metrics exported here can be used.

Please comment if you are interested in helping to maintain this module :) Thanks!

https://github.com/prometheus/client_java/releases/tag/v1.0.0

Hey Team http4s - hope you are well.

Is there any reason why the status label isn't added when observing the response time - https://github.com/http4s/http4s-prometheus-metrics/blob/series/0.24/prometheus-metrics/src/main/scala/org/http4s/metrics/prometheus/Prometheus.scala#L136-L138?

I'm guessing it's not recorded since the recordHeaders usage of observe would not have it?

Thanks