#!/usr/bin/env stack
– stack –install-ghc runghc –package monad-metrics –package turtle –package markdown-unlit – “-pgmL markdown-unlit”
monad-metrics
This library defines a convenient wrapper and API for using EKG metrics in
your application. It’s heavily inspired by the metrics code that Taylor Fausak
used in his Haskell application blunt.
Usage
This README is an executable literate Haskell
file. If you have stack installed, then you can run the file with:
./README.lhs
We’ll need to start with the import/pragma boilerplate:
{-# LANGUAGE OverloadedStrings #-}
{-# LANGUAGE FlexibleInstances #-}
{-# LANGUAGE NoMonomorphismRestriction #-}
import qualified Control.Monad.Metrics as Metrics
import Control.Monad.Metrics (Metrics, Resolution(..), MonadMetrics(..))
import Control.Monad.Reader
import qualified System.Metrics as EKG
The Control.Monad.Metrics
module is designed to be imported qualified.
Initialize!
First, you need to initialize the Metrics
data type. You can do so using
initialize
(to create a new EKG store) or initializeWith
if you want to
pass a preexisting store.
initializing :: Bool -> EKG.Store -> IO Metrics
initializing True store = Metrics.initializeWith store
initializing False _ = Metrics.initialize
Embed!
The next step is to implement an instance of the class MonadMetrics
for your
monad transformer stack. This library has explicitly decided not to provide a
concrete monad transformer to reduce the dependency footprint. Fortunately,
it’s pretty easy!
Suppose you’ve got the following stack:
type App = ReaderT Config IO
data Config = Config { configMetrics :: Metrics }
then you can easily get the required instance with:
instance MonadMetrics (ReaderT Config IO) where
getMetrics = asks configMetrics
Now, you’re off to the races! Let’s record some metrics.
If you’re after a really simple embedding, you can use run
or run'
:
simple :: Int -> IO ()
simple i =
Metrics.run $ do
metrics <- Metrics.getMetrics
Metrics.gauge "Simple" i
forM_ [1..i] $ \_ -> do
Metrics.increment "Count!"
gettingThere :: IO ()
gettingThere =
Metrics.run' (\metrics -> Config metrics) $ do
liftIO $ putStrLn "it accepts a constructor"
Measure!
Once your application has the required instance, you can use EKG’s metrics
(counters, gauges, labels, distributions).
For detailed descriptions of the various metric types, see the corresponding EKG documentation:
Generally, the library provides “sane default” functions which accept the name
of the metric to work with and the value to contribute to that metric.
w = Metrics.label "Foo" "Bar"
x = Metrics.counter "MetricName" 6
y = Metrics.distribution "Distribute" 3.4
z = Metrics.gauge "Gauge" 7
Generalized versions of these functions are available with an apostrophe. Labels accept any Show
able value, while gauges and counters accept any Integral
value.
a = Metrics.label' "List" [1,2,3]
b = Metrics.counter' "Count" (3 :: Integer)
Timers
You can time actions with timed
, which has a resolution of seconds. You can
use timed'
which accepts a Resolution
argument to provide a different
scale.
timedProcess :: App Int
timedProcess =
Metrics.timed "summing1" $ do
pure $! sum [1 .. 100000]
timedInMilliseconds :: App Int
timedInMilliseconds =
Metrics.timed' Microseconds "summing2" $ do
pure $! sum [1..100]
A demonstration
main :: IO ()
main = do
metrics <- Metrics.initialize
flip runReaderT (Config metrics) $ do
Metrics.label "ProgramName" "README"
forM_ [1..10] $ \_ -> do
Metrics.increment "up-to-ten"
Metrics.timed' Nanoseconds "Whatever" $ do
liftIO $ putStrLn "Hello World!"