Instrumenting Servant with Prometheus

To set up a Servant/WAI application with monitoring, we will first scaffold a servant application using stack.

> stack new servant-prometheus \ \
--resolver nightly-2016-05-21

To instrument our servant application, add the following three dependencies to servant-prometheus.cabal under the build-depends of the library.

, wai-middleware-prometheus >=
, prometheus-client >=
, prometheus-metrics-ghc >=

In src/Lib.hs we need the following 3 imports. They are fully qualified for clarity.

import Network.Wai.Middleware.Prometheus (prometheus, PrometheusSettings(..))
import Prometheus (register)
import Prometheus.Metric.GHC (ghcMetrics)

All of our work will be in the startApp function defined in the same file. Currently, startApp is defined as run 8080 app. We are going to change that to the following.

startApp = do
  register ghcMetrics
  let promMiddleware = prometheus $ PrometheusSettings ["metrics"] True True
  run 8080 $ promMiddleware $ app

We also need {-# LANGUAGE OverloadedStrings #-} for the ["metrics"] literal.

Finally, to actually enable consumption of GHC metrics, we need to include the -T flag in our executable. The executable's ghc-options becomes:

ghc-options: -threaded -rtsopts -with-rtsopts=-TN

After building and running, the servant server should be providing prometheus metrics at /metrics.

stack install --only-dependencies
stack build
stack exec servant-prometheus-exe


All three of the packages we added as dependencies come from fimad/prometheus-haskell.

  • Prometheus Client
    • Defines the core data types and metrics. The other two packages build on client to provide their metrics.
    • This is the package to use to set up custom metrics.
  • Prometheus Metrics GHC
    • Provides a set of custom metrics for Prometheus Client that are exposed from GHC’s runtime system.
  • WAI Middleware Prometheus
    • Instruments a WAI application, exposing common HTTP metrics. Also exposes those metrics on /metrics (or the url of the user’s choice)

localhost:8080/metrics should now have the following output, which can be scraped by Prometheus.

  • http_ prefixed metrics come from the WAI middleware
  • ghc_ prefixed metrics come from ghcMetrics
# HELP http_request_duration_microseconds The HTTP request latencies in microseconds.
# TYPE http_request_duration_microseconds summary
http_request_duration_microseconds{handler="app",quantile="0.5"} 258.0
http_request_duration_microseconds{handler="app",quantile="0.9"} 258.0
http_request_duration_microseconds{handler="app",quantile="0.99"} 258.0
http_request_duration_microseconds_sum{handler="app"} 258.0
http_request_duration_microseconds_count{handler="app"} 1
# HELP ghc_sparks The number of sparks in the local spark pool.
# TYPE ghc_sparks gauge
ghc_sparks 0
# HELP ghc_capabilities The number of threads that can run truly simultaneously.
# TYPE ghc_capabilities gauge
ghc_capabilities 1
# HELP ghc_bytes_allocated Total number of bytes allocated.
# TYPE ghc_bytes_allocated gauge
ghc_bytes_allocated 249136
# HELP ghc_num_gcs The number of garbage collections performed.
# TYPE ghc_num_gcs gauge
ghc_num_gcs 3
# HELP ghc_max_bytes_used The maximum number of live bytes seen so far.
# TYPE ghc_max_bytes_used gauge
ghc_max_bytes_used 65576
# HELP ghc_cumulative_bytes_used The cumulative total bytes used.
# TYPE ghc_cumulative_bytes_used gauge
ghc_cumulative_bytes_used 149136
# HELP ghc_bytes_copied The number of bytes copied during garbage collection.
# TYPE ghc_bytes_copied gauge
ghc_bytes_copied 76344
# HELP ghc_current_bytes_used The number of current live bytes.
# TYPE ghc_current_bytes_used gauge
ghc_current_bytes_used 65504
# HELP ghc_current_bytes_slop The current number of bytes lost to slop.
# TYPE ghc_current_bytes_slop gauge
ghc_current_bytes_slop 0
# HELP ghc_max_bytes_slop The maximum number of bytes lost to slop so far.
# TYPE ghc_max_bytes_slop gauge
ghc_max_bytes_slop 8152
# HELP ghc_peak_megabytes_allocated The maximum number of megabytes allocated.
# TYPE ghc_peak_megabytes_allocated gauge
ghc_peak_megabytes_allocated 1
# HELP ghc_mutator_cpu_seconds The CPU time spent running mutator threads.
# TYPE ghc_mutator_cpu_seconds gauge
ghc_mutator_cpu_seconds 1.471e-3
# HELP ghc_mutator_wall_seconds The wall clock time spent running mutator threads.
# TYPE ghc_mutator_wall_seconds gauge
ghc_mutator_wall_seconds 1.48e-3
# HELP ghc_gc_cpu_seconds The CPU time spent running GC.
# TYPE ghc_gc_cpu_seconds gauge
ghc_gc_cpu_seconds 1.428e-3
# HELP ghc_gc_wall_seconds The wall clock time spent running GC.
# TYPE ghc_gc_wall_seconds gauge
ghc_gc_wall_seconds 1.781416e-3
# HELP ghc_cpu_seconds Total CPU time elapsed since program start.
# TYPE ghc_cpu_seconds gauge
ghc_cpu_seconds 0.352497
# HELP ghc_wall_seconds Total wall clock time elapsed since start.
# TYPE ghc_wall_seconds gauge
ghc_wall_seconds 28.681236189
# HELP ghc_parallel_total_bytes_copied Number of bytes copied during GC, minus space held by mutable lists held by the capabilities.
# TYPE ghc_parallel_total_bytes_copied gauge
ghc_parallel_total_bytes_copied 0
# HELP ghc_parallel_max_bytes_copied Sum of number of bytes copied
each GC by the most active GC thread each GC.
# TYPE ghc_parallel_max_bytes_copied gauge
ghc_parallel_max_bytes_copied 0