Blammo

Blammo is a Structured Logging library that's

• Easy to use: one import and go!
• Easy to configure: environment variable parsing out of the box!
• Easy to integrate: see below for Amazonka, Yesod, and more!
• Produces beautiful, colorful output in development
• Produces fast-fast JSON in production

All built on the well-known MonadLogger interface and using an efficient fast-logger implementation.

It's better than bad, it's good!

Simple Usage

import Blammo.Logging.Simple

Throughout your application, you should write against the ubiquitous MonadLogger interface:

action1 :: MonadLogger m => m ()
action1 = do
  logInfo "This is a message sans details"

And make use of monad-logger-aeson for structured details:

data MyError = MyError
  { code :: Int
  , messages :: [Text]
  }
  deriving stock Generic
  deriving anyclass ToJSON

action2 :: MonadLogger m => m ()
action2 = do
  logError $ "Something went wrong" :# ["error" .= MyError 100 ["x", "y"]]
  logDebug "This won't be seen in default settings"

When you run your transformer stack, wrap it in runLoggerLoggingT providing any value with a HasLogger instance (such as your main App). The Logger type itself has such an instance, and we provide runSimpleLoggingT for the simplest case: it creates one configured via environment variables and then calls runLoggerLoggingT with it.

You can use withThreadContext (from monad-logger-aeson) to add details that will appear in all the logged messages within that scope. Placing one of these at the very top-level adds details to all logged messages.

runner :: LoggingT IO a -> IO a
runner = runSimpleLoggingT . withThreadContext ["app" .= ("example" :: Text)]

main :: IO ()
main = runner $ do
  action1
  action2

The defaults are good for CLI applications, producing colorful output (if connected to a terminal device) suitable for a human:

Under the hood, Logging.Settings.Env is using envparse to configure logging through environment variables. See that module for full details. One thing we can adjust is LOG_LEVEL:

In production, you will probably want to set LOG_FORMAT=json and ship logs to some aggregator like Datadog or Mezmo (formerly LogDNA):

Multiline Format

With the terminal formatter, a log message that is more than 120 visible characters will break into multi-line format:

This breakpoint can be controlled with LOG_BREAKPOINT. Set an unreasonably large number to disable this feature.

Out of Order Messages

Blammo is built on fast-logger, which offers concurrent logging through multiple buffers. This concurrent logging is fast, but may deliver messages out of order. You want this on production: your aggregator should be inspecting the message's time-stamp to re-order as necessary on the other side. However, this can be problematic in a CLI, where there is both little need for such high performance and a lower tolerance for the confusion of out of order messages.

For this reason, the default behavior is to not use concurrent logging, but setting the format to json will automatically enable it (with {number-of-cores} as the value). To handle this explicitly, set LOG_CONCURRENCY.

Configuration

Setting	Setter	Environment variable and format
Format	setLogSettingsFormat	LOG_FORMAT=tty|json
Level(s)	setLogSettingsLevels	LOG_LEVEL=<level>[,<source:level>,...]
Destination	setLogSettingsDestination	LOG_DESTINATION=stdout|stderr|@<path>
Color	setLogSettingsColor	LOG_COLOR=auto|always|never
Breakpoint	setLogSettingsBreakpoint	LOG_BREAKPOINT=<number>
Concurrency	setLogSettingsConcurrency	LOG_CONCURRENCY=<number>

Advanced Usage

Add our environment variable parser to your own,

data AppSettings = AppSettings
  { appDryRun :: Bool
  , appLogSettings :: LogSettings
  , -- ...
  }

loadAppSettings :: IO AppSettings
loadAppSettings = Env.parse id $ AppSettings
  <$> var switch "DRY_RUN" mempty
  <*> LogSettingsEnv.parser
  <*> -- ...

Load a Logger into your App type and define HasLogger,

data App = App
  { appSettings :: AppSettings
  , appLogger :: Logger
  , -- ...
  }

instance HasLogger App where
  loggerL = lens appLogger $ \x y -> x { appLogger = y }

loadApp :: IO App
loadApp = do
  appSettings <- loadAppSettings
  appLogger <- newLogger $ appLogSettings appSettings
  -- ...
  pure App {..}

Use runLoggerLoggingT,

runAppT :: App -> ReaderT App (LoggingT IO) a -> IO a
runAppT app f = runLoggerLoggingT app $ runReaderT f app

Use without LoggingT

If your app monad is not a transformer stack containing LoggingT (ex: the ReaderT pattern), you can implement a custom instance of MonadLogger:

data AppEnv = AppEnv
  { appLogFunc :: Loc -> LogSource -> LogLevel -> LogStr -> IO ()
  -- ...
  }

newtype App a = App
  { unApp :: ReaderT AppEnv IO a }
  deriving newtype
    ( Functor
    , Applicative
    , Monad
    , MonadIO
    , MonadReader AppEnv
    )

instance MonadLogger App where
  monadLoggerLog loc logSource logLevel msg = do
    logFunc <- asks appLogFunc
    liftIO $ logFunc loc logSource logLevel (toLogStr msg)

runApp :: AppEnv -> App a -> IO a
runApp env action =
  runReaderT (unApp action) env

In your app you can use code written against the MonadLogger interface, like the actions defined earlier:

app :: App ()
app = do
  action1
  action2

To retrieve the log function from Blammo, use askLoggerIO (from MonadLoggerIO) with runSimpleLoggingT (or runLoggerLoggingT if you need more customization options), when you initialize the app:

main2 :: IO ()
main2 = do
  logFunc <- runSimpleLoggingT askLoggerIO
  let appEnv =
        AppEnv
          { appLogFunc = logFunc
          -- ...
          }
  runApp appEnv app

Integration with RIO

data App = App
  { appLogFunc :: LogFunc
  , -- ...
  }

instance HasLogFuncApp where
  logFuncL = lens appLogFunc $ \x y -> x { logFunc = y }

runApp :: MonadIO m => RIO App a -> m a
runApp f = runSimpleLoggingT $ do
  loggerIO <- askLoggerIO

  let
    logFunc = mkLogFunc $ \cs source level msg -> loggerIO
      (callStackLoc cs)
      source
      (fromRIOLevel level)
      (getUtf8Builder msg)

  app <- App logFunc
    <$> -- ...
    <*> -- ...

  runRIO app $ f

callStackLoc :: CallStack -> Loc
callStackLoc = undefined

fromRIOLevel :: RIO.LogLevel -> LogLevel
fromRIOLevel = undefined

Integration with Amazonka

data App = App
  { appLogger :: Logger
  , appAWS :: AWS.Env
  }

instance HasLogger App where
  -- ...

runApp :: ReaderT App (LoggingT IO) a -> IO a
runApp f = do
  logger <- newLogger defaultLogSettings
  app <- App logger <$> runLoggerLoggingT logger awsDiscover
  runLoggerLoggingT app $ runReaderT f app

awsDiscover :: (MonadIO m, MonadLoggerIO m) => m AWS.Env
awsDiscover = do
  loggerIO <- askLoggerIO
  env <- liftIO $ AWS.newEnv AWS.discover
  pure $ env
    { AWS.envLogger = \level msg -> do
      loggerIO
        defaultLoc -- TODO: there may be a way to get a CallStack/Loc
        "Amazonka"
        (case level of
          AWS.Info -> LevelInfo
          AWS.Error -> LevelError
          AWS.Debug -> LevelDebug
          AWS.Trace -> LevelOther "trace"
        )
        (toLogStr msg)
    }

Integration with WAI

import Network.Wai.Middleware.Logging

instance HasLogger App where
  -- ...

waiMiddleware :: App -> Middleware
waiMiddleware app =
  addThreadContext ["app" .= ("my-app" :: Text)]
    $ requestLogger app
    $ defaultMiddlewaresNoLogging

Integration with Warp

instance HasLogger App where
  -- ...

warpSettings :: App -> Settings
warpSettings app = setOnException onEx $ defaultSettings
 where
  onEx _req ex =
    when (defaultShouldDisplayException ex)
      $ runLoggerLoggingT app
      $ logError
      $ "Warp exception"
      :# ["exception" .= displayException ex]

Integration with Yesod

instance HasLogger App where
  -- ...

instance Yesod App where
  -- ...

  messageLoggerSource app _logger loc source level msg =
    runLoggerLoggingT app $ monadLoggerLog loc source level msg

---

LICENSE | CHANGELOG