glue

Building Better Services And Clients

Monthly Downloads: 0

Programming language: Haskell

License: BSD 3-clause "New" or "Revised" License

Tags: Network     Glue    

README

Building Better Services And Clients

This package provides methods to deal with the common needs of services talking to other services, like timeouts and retries, treating these as cross-cutting concerns that aren't tied to a specific transport like HTTP.

Examples

Batching

In a social network there may be another service that returns user records, if this social network is really busy it would likely be more efficient to capture multiple calls and dispatch them as one multi-get call.

The batchingService function creates both single and multi-get calls (hence the "fmap snd" below), which accumulate requests over a user defined window in time and dispatch them once that window has passed. The service functions created can be used across threads, thereby limiting the amount of parallel calls to an upstream system.

data User = User Int String

makeUser :: Int -> User
makeUser userId = User userId ("User " ++ (show userId))

userService :: S.HashSet Int -> IO (M.HashMap Int User)
userService request = do
  threadDelay 500
  return $ M.fromList $ fmap (\r -> (r, makeUser r)) $ S.toList request

batchedUserService :: IO (S.HashSet Int -> IO (M.HashMap Int User))
batchedUserService = fmap snd $ batchingService defaultBatchingOptions userService

Retries

Often we want to retry service calls if they've failed because it might've been a transient error that subsequently succeeds.

failingService :: IORef Int -> Int -> IO Int
failingService ref request = do
  counter <- atomicModifyIORef' ref (\c -> (c + 1, c + 1))
  if counter `mod` 3 == 0 then fail "Bang!" else return (request * 2)

notSoFailingService :: IO (Int -> IO Int)
notSoFailingService = do
  ref <- liftIO $ newIORef 0
  return $ retryingService defaultRetryOptions $ failingService ref

Caching

serviceThatNeedsCaching :: Int -> IO Int
serviceThatNeedsCaching request = do
  putStrLn "Doing Something Expensive!"
  return (request * 2)

cachedService :: IO (Int -> IO Int)
cachedService = do
  cache <- newAtomicLRU $ Just 100
  return $ cacheWithLRU cache serviceThatNeedsCaching

Todo

• Load balancer, maybe just a simple round robin to start with.
• Concurrent request constraint, possibly X inflight and Y queued.
• Routing service? Possibly something akin to HTTP routing in a lot of web frameworks.
• Request duplication, with completion based on the first successful result.
• Graceful service shutdown, reject new requests with a default value, wait for a period of time while existing requests continue. Then possibly a cancel action?