we use flink metrics to expose custom metrics -

<https://nightlies.apache.org/flink/flink-docs-master/docs/ops/metrics/>

but we have no idea how to add custom labels for metrics. even it’s possible or not, or what’s best practice for it in example:

class MyMapper extends RichMapFunction[String,String] {
  @transient private var counter: Counter = _

  override def open(parameters: Configuration): Unit = {
    counter = getRuntimeContext()
      .getMetricGroup()
      .counter("myCounter")
  }

  override def map(value: String): String = {
    value match {
                  case "OK" => counter.inc("customLabel: SUCCESS") 
                  case "KO" => counter.inc("customLabel: FAIL") 
                  case _ => counter.inc("customLabel: UNKNOWN") 
                }
    value
  }
}

so depends on data we want to add some labels Is that possible? crazy idea comes to my head about create side output, generate metrics there and sink them via httpSink to prometheus push gateway 😄

Labels are just more groups in metric 😊 as far as I understand

getRuntimeContext()
          .getMetricGroup()
          .addGroup("valid", result.toString)
          .addGroup("event", event)
          .addGroup("team", team)
          .counter("count")

Issue here that you’re not allowed to put custom metric depends on logic inside

processFunction

it’s fully static thing, and we want to make it a bit dynamic

That’s exactly what I’m doing with this code

wait, you have following code

metric = getRuntimeContext()
          .getMetricGroup()
          .addGroup("valid", result.toString)
          .addGroup("event", event)
          .addGroup("team", team)
          .counter("count")
metric.inc()

inside process function, not in

open()

?

In open I define

_val nameToCounter_: mutable.HashMap[String, _Counter_]

When I first get specific set of values for this metric I put this counter into hashMap and use this HashMap in .process()

i have not been able to get grouping to act as a label, at least for prometheus. i wind up with string values concatenated into the metric name.