Reading at scrape time · metrics4s · Sendline Open Source

Reading at scrape time

Some values are cheaper to read when Prometheus asks for them. Each of these gives an Allocated[F, Unit], which the runtime turns into its own resource (see Runtimes):

scala

Metrics[IO].observe(cacheSize)(_.report(cache.size.toDouble))

Metrics[IO].observeAll(scheduler.status())(                  // one read per scrape, several metrics
  Observation(workers)((status, out) => out.report(status.workers.toDouble)),
  Observation(load)((status, out) => out.report(status.load)),
)

Metrics[IO].observeAllF(queue.size, timeout = 1.second)(      // a value only available in F
  Observation(queueSize)((size, out) => out.report(size.toDouble))
)

A callback that throws, or an F that runs past its timeout, does not break the scrape. Its metrics are left out of that scrape, the F is cancelled, and metrics4s_callback_errors_total{metric="<first metric>"} counts it.