Juspay's EulerHS Framework

EulerHS is a free monadic framework for easy building concurrent backend and console applications in Haskell. This framework provides you with the most important subsystems already integrated, such as SQL DBs, logging, KV DBs and other.

The framework represents a safe layer with its own philosophy of exception safety, so you don't need to think about those difficult Haskell questions.

The framework also offers two testing mechanisms: integration testing facilities and automatic whitebox testing mechanism. The code you'll be writing will be simple and testable. See the related materials for more info.

Framework is successfully used in production in Juspay and shows impressive results.

• Framework features
• Business logic code sample
• Important note on forked external libs
  • hedis library for Redis
  • beam-mysql library for MySql
  • beam
• EulerHS application architecture
• Flow monad sample usages
  • Logging
  • Typed options
  • SQL subsystem
  • mtl style support
  • KV DB subsystem
  • Concurrency
  • Forking and awaiting child flows
• Building the framework
• Testing the framework
• EulerHS tutorials and template projects
• Background materials

Framework features

The framework exports the Flow monad which provides the following facilities:

• Custom Prelude (universum-based)
• SQL DB interaction using the beam library. Supported SQL backends:
  • Postgres
  • MySQL
  • SQLite
• KV DB interaction. Supported KV DBs:
  • Redis
• Async and parallel flows evaluation
• Client HTTP in two forms:
  • Servant-based HTTP client interaction
  • Low-level HTTP client
• Logging (tiny-logger inside)
• Typed mutable options
• Safe IO actions
• Exception-handling mechanism
• Running OS system commands
• ART (Automatic Regression Testing) - white box testing facilities
• Integration testing framework
• Experimental Pub/Sub mechanism (using Redis pub sub subsystem)

Business logic code sample

A sample scenario you may find here.

import           EulerHS.Prelude
import qualified EulerHS.Language as L
import           EulerHS.TestData.API.Client
import           EulerHS.TestData.Types
import           Servant.Client (BaseUrl (..), Scheme (..))

testScenario1 :: L.Flow User
testScenario1 = do
  logDebug "testScenario1" "Running sys cmd whoami..."
  localUserName <- L.runSysCmd "whoami"

  logDebug "testScenario1" "Reading a guid from file..."
  localGUID <- L.runIO (readFile "my_guid.txt")

  logDebug "testScenario1" "Generating new guid..."
  guid <- L.generateGUID

  logDebug "testScenario1" "Obtaining URL..."
  url <- maybe (mkUrl "127.0.0.1") mkUrl <$> L.getOption UrlKey

  logDebug "testScenario1" "Calling some HTTP API..."
  res <- L.callServantAPI Nothing url getUser

  logDebug "testScenario1" "Finished."
  pure $ case res of
    Right u | localGUID /= userGUID u -> u
    Right u | otherwise -> User localUserName "" $ toString guid
    _ -> User localUserName "Smith" $ toString guid
  where
    mkUrl :: String -> BaseUrl
    mkUrl host = BaseUrl Http host port ""

Important note on forked external libs

Juspay made forks of several libraries to fix their problems and to support specific cases. You might want to be aware of these Juspay-specific fixes. If they are not suitable for your needs, it might need to avoid them till EulerHS v3.0.

hedis library for Redis

hedis is stapled to a Juspay-specific fork as current. This fork is multiple releases behind the current mainline, incompatibly.

beam-mysql library for MySql

The beam-mysql is rewritten almost completely. The original version doesn't have protection from SQL injections, and also is written with some internal problems. The updated version fixes that.

More info on beam usage can be found in BEAM-NOTES.

beam

We made several minor improvements of the original beam library in our fork. These changes do not have anything Juspay-specific, but yet to be pushed to the upstream.

More info on beam usage can be found in BEAM-NOTES.

EulerHS application architecture

The EulerHS framework slices your application into several layers preventing the implementation details to leak into the business logic. This helps to keep the business logic code simple, maintainable and more reliable. It also helps to tweak the implementation without affecting the business logic. The further development of the framework will be towards improving its safety and performance, but this won't require updating of the business logic code due to this clear layering.

Layers of the typical web backend application with Servant:

Application layer:

• Application (IO monad)
  • Processes app configs and command line arguments
  • Manages FlowRuntime
  • Runs Servant HTTP server
• API types model
  • Types used in Servant to handle queries and API methods
• Servant method handler (Servant monad stack: ReaderT FlowRuntime (ExceptT ServerError IO))
  • Handles errors and exceptions
  • Validates requests
  • Converts API types into domain types and back
  • Uses API types model
  • Runs business logic scenarios using FlowRuntime and interpreters

Business domain layer:

• EulerHS language (the Flow monad and its derivables)
  • Provides an abstracted, purified set of eDSLs which are free from implementation details
  • Helps to organize business logic code
• Domain model (ADT types mostly)
  • Represents types and functions directly related to the domain
• DB model (beam-powered schema)
  • DB representation of domain
• Business logic scenarios (Flow monad and its derivables)
  • Interacts with domain model and DB model
  • Isolated from the other layers

Implementation layer:

• EulerHS runtime (FlowRuntime)
  • Keeps operational data of the framework
  • Handles resources, manages connections
• EulerHS interpreters (IO-based monadic stack)
  • Connects Flow scenarios to real subsytems and libraries
  • Handles exceptions, threads, ART system
  • Uses FlowRuntime

Checkout Background materials to know more about this layering.

Flow monad sample usages

Logging

Framework provides logging mechanism out-of-the-box. It provides 4 logging functions:

logInfo    :: Show tag => tag -> Message -> Flow ()
logError   :: Show tag => tag -> Message -> Flow ()
logDebug   :: Show tag => tag -> Message -> Flow ()
logWarning :: Show tag => tag -> Message -> Flow ()

Usage is quite simple.

import qualified EulerHS.Language as L

myFlow :: L.Flow ()
myFlow = L.logInfo "myFlow" "Hello world!"

Notice there is no anything related to a specific logging library. It's all hidden behind the Flow interface.

The logging subsystem can be configured on the start of the application. You should specify what logger you want passing a logger creation function into withFlowRuntime or createFlowRuntime functions.

• No logger (createVoidLoggerRuntime)
• Memory logger (createMemoryLoggerRuntime). Can eat your memory very fast!
• File logger (a special flag in LoggerConfig). Will flush logs into a file (the action is not immediate).
• Console logger (a special flag in LoggerConfig). Will show logs in console.

You can also choose should your file and console logger be sync or async (a special flag in LoggerConfig).

N.B. Async logger is not properly tested yet, it can have performance implications.

import qualified EulerHS.Types as T
import qualified EulerHS.Runtime as R
import qualified EulerHS.Interpreters as I

loggerConfig :: T.LoggerConfig
loggerConfig = T.LoggerConfig
    { T._isAsync = False
    , T._logLevel = Debug
    , T._logFilePath = "/tmp/logs/myFlow.log"
    , T._logToConsole = True
    , T._logToFile = True
    , T._maxQueueSize = 1000
    , T._logRawSql = False
    }

runApp :: IO ()
runApp = do
  let mkLoggerRt = R.createLoggerRuntime T.defaultFlowFormatter loggerConfig
  R.withFlowRuntime (Just mkLoggerRt)
    $ \flowRt -> I.runFlow flowRt myFlow

The framework also supports different logger formatters for different flows. See documentation on FlowFormatter for more info.

On FlowRuntime disposal, the rest of the queue of logs will be flushed gracefully.

Log entries are considered app-wide, but there is a LogCounter that is counting entries and helping to see their ordering.

Ordering itself is not guaranteed.

Typed options

Just typed key-value options.

getOption :: (OptionEntity k v) => k -> Flow (Maybe v)
setOption :: (OptionEntity k v) => k -> v -> Flow ()
delOption :: (OptionEntity k v) => k -> Flow ()

Options work as a shared concurrent yet mutable state, so be careful to not produce data races.

Avoid using it as an operational state of your app, it's better to use StateT on top of the Flow. See State handling for more info.

data TestIntKey = TestIntKey
  deriving (Generic, Typeable, Show, Eq, ToJSON, FromJSON)

data TestStringKey = TestStringKey
  deriving (Generic, Typeable, Show, Eq, ToJSON, FromJSON)

instance T.OptionEntity TestStringKey String
instance T.OptionEntity TestIntKey Int

myFlow :: L.Flow (Maybe String, Maybe Int)
myFlow = do
  _  <- L.setOption TestStringKey "lore ipsum"
  _  <- L.setOption TestIntKey 100
  v1 <- L.getOption TestStringKey
  v2 <- L.getOption TestIntKey
  pure (v1, v2)

-- (Just "lore ipsum", Just 100)

SQL subsystem

Supported SQL backends

This subsystem supports several SQL backends:

• MySQL
• Postgres
• SQLite

The goal was to provide a unified interface for all those backends, so that different projects could have a relatively similar code.

Unfortunately, there are two drawbacks here.

• We had to fix many problems in beam and related libraries, but were unable to push all the changes to the upstream. This especially true regarding beam-mysql that we had to rewrite almost completely.
• Framework cannot provide things specific to any SQL backend. Only a common subset of features.
• beam itself is a sophisticated library, and its usage is quite difficult. You may want to consult with our cheatsheet on beam usage.

Connectivity and pools

Framework allows you to create either permanent, application-wide SQL connections, or immediate single-usage connections in place. The Flow language provides two methods for connecting and one for disconnecting:

initSqlDBConnection    :: DBConfig beM -> Flow (DBResult (SqlConn beM))
getOrInitSqlConnection :: DBConfig beM -> Flow (DBResult (SqlConn beM))
deinitSqlDBConnection  :: SqlConn beM  -> Flow ()

All the connections are held in pools by design (resource-pool library). There is no way to avoid this (except maybe making your own connections with runIO). You need to fill the DBConfig structure to setup the pool.

import EulerHS.Types as T
import qualified Database.Beam.MySQL as BM

poolConfig = T.PoolConfig
  { stripes = 1
  , keepAlive = 10
  , resourcesPerStripe = 50
  }

sqliteCfg :: DBConfig BM.SqliteM
sqliteCfg = T.mkSQLitePoolConfig "SQliteDB" testDBName poolConfig

Here, SqlietM is a phantom type provided by the beam-mysql library. Every SQL backend has own phantom type to distinguish between them. You can use smart constructors for producing configs for every backend:

import "beam-mysql"    Database.Beam.MySQL as BM
import "beam-postgres" Database.Beam.Postgres as BP
import "beam-sqlite"   Database.Beam.Sqlite as BS

mkSQLitePoolConfig   :: ConnTag -> SQliteDBname   -> PoolConfig -> DBConfig BM.SqliteM
mkPostgresPoolConfig :: ConnTag -> PostgresConfig -> PoolConfig -> DBConfig BP.Pg
mkMySQLPoolConfig    :: ConnTag -> MySQLConfig    -> PoolConfig -> DBConfig BS.MySQLM

The ConnTag parameter is a Text mark that is used to uniquely identify a pool of your connections. If you want to have several pools (for example, for several distinct SQL backends), you should provide unique tags for each pool.

Simplified smart constructors use a default pool setting:

mkSQLiteConfig   :: ConnTag -> SQliteDBname   -> DBConfig BS.SqliteM
mkPostgresConfig :: ConnTag -> PostgresConfig -> DBConfig BP.Pg
mkMySQLConfig    :: ConnTag -> MySQLConfig    -> DBConfig BM.MySQLM

defaultPoolConfig :: PoolConfig
defaultPoolConfig = PoolConfig
  { stripes = 1
  , keepAlive = 100
  , resourcesPerStripe = 1
  }

DB model and DB schema with beam

In order to query your database, you should define a DB model for its schema. This model will encode your tables and relations according to beam requirements.

import qualified Database.Beam as B

-- Description of the @member@ table

data MemberT f = Member
    { memberId      :: B.C f Int
    , surName       :: B.C f Text
    , firstName     :: B.C f Text
    , address       :: B.C f Text
    , zipCode       :: B.C f Int
    , telephone     :: B.C f Text
    , recommendedBy :: B.C f (Maybe Int)
    , joinDate      :: B.C f LocalTime
    } deriving (Generic, B.Beamable)

-- Description of the primary key type:

instance B.Table MemberT where
  data PrimaryKey MemberT f = MemberId (B.C f Int)
    deriving (Generic, B.Beamable)
  primaryKey = MemberId . memberId

type Member = MemberT Identity
type MemberId = B.PrimaryKey MemberT Identity

-- Field names can be mapped 1:1 to names of the ADT, but it's possible
-- to alter them, like this:

membersEMod
  :: B.EntityModification (B.DatabaseEntity be db) be (B.TableEntity MemberT)
membersEMod = B.modifyTableFields
  B.tableModification
    { memberId      = B.fieldNamed "memid"
    , surName       = B.fieldNamed "surname"
    , firstName     = B.fieldNamed "firstname"
    , address       = B.fieldNamed "address"
    , zipCode       = B.fieldNamed "zipcode"
    , telephone     = B.fieldNamed "telephone"
    , recommendedBy = B.fieldNamed "recommendedby"
    , joinDate      = B.fieldNamed "joindate"
    }

-- The Schema itself:

data ClubDB f = ClubDB
    { members    :: f (B.TableEntity MemberT)   -- members table
    , facilities :: f (B.TableEntity FacilityT) -- facilities table
    , bookings   :: f (B.TableEntity BookingT)  -- bookings table
    } deriving (Generic, B.Database be)

-- DB Schema representation. Use this value to reference to any of tables.
clubDB :: B.DatabaseSettings be ClubDB
clubDB = B.defaultDbSettings `B.withDbModification` B.dbModification
    { facilities = facilitiesEMod     -- Alter field names of tables.
    , members    = membersEMod
    , bookings   = bookingsEMod
    }

There can be conversion functions between your domain model and DB models, but the simplest approach would be to stick with only the DB model.

Querying

A typical SELECT query for the members table is presented in the following code. It requests all members joined after this date:

searchByDate :: LocalTime -> L.Flow (T.DBResult [Member])
searchByDate startDate = do
  conn <- connectOrFail sqliteCfg       -- obtain SQLite config somehow
  L.runDB conn                          -- run a query non-transactionally
    $ L.findRows                        -- SELECT query, returns rows
    $ B.select
    $ B.filter_ (\m -> joinDate m >=. B.val_ startDate)
    $ B.all_ (members clubDB)

Notice that we use runDB for running SQL queries expressed in beam and wrapped into the SqlDB language. The runDB evaluates the DB script non-transactionally. Use another version of this function, runDBTransaction instead. In this case, any query packed into an SqlDB monadic block, will be scoped by a single transaction.

beam exposes a set of methods to construct insert, update and delete queries. These all are integrated into the framework. Consult with this test suite QueryExamplesSpec on how to compose queries with beam and SqlDB.

Permanent connections

For permanent connections, you need to create them on the Application Layer and pass them into your Flow scenarios. One of the possible solutions will be to wrap your Flow scenarios into a ReaderT stack (so called ReaderT pattern), and provide an environment with permanent connections:

import qualified Database.Beam.Sqlite as BS

data FlowEnv = FlowEnv
  { sqliteConn1 :: T.SqlConn BS.SqliteM
  , sqliteConn2 :: T.SqlConn BS.SqliteM
  }

type MyFlow a = ReaderT FlowEnv L.Flow a
-- The same as
-- type MyFlow r a = L.ReaderFlow FlowEnv a

searchByDate :: LocalTime -> MyFlow (T.DBResult [Member])
searchByDate startDate = do
  FlowEnv conn1 _ <- ask
  L.runDB conn1                         -- run a query
    $ L.findRows                        -- SELECT query returning rows
    $ B.select
    $ B.filter_ (\m -> joinDate m >=. B.val_ startDate)
    $ B.all_ (members clubDB)

Here, DBResult is an Either type that carries either success or a failure:

data DBError = DBError DBErrorType Text
  deriving (Show, Eq, Ord, Generic, ToJSON, FromJSON)

type DBResult a = Either DBError a

Where DBErrorType is an enum with various reasons of DB failure.

For more info on the SQL DB subsystem usage, see tutorials and background materials.

mtl style support

The framework exposes a type class MonadFlow with instances for Flow, ReaderT r Flow, StateT s Flow, WriterT w Flow, ExceptT e Flow and some other transformers. This makes it possible to write your Flow scenarios with the mtl style.

Let's see how the scenario searchByDate will look like:

searchByDate
  :: MonadFlow m
  => MonadReader FlowEnv m
  => LocalTime
  -> m (T.DBResult [Member])
searchByDate startDate = do
  FlowEnv conn1 _ <- ask
  L.runDB conn1                         -- run a query
    $ L.findRows                        -- SELECT query returning rows
    $ B.select
    $ B.filter_ (\m -> joinDate m >=. B.val_ startDate)
    $ B.all_ (members clubDB)

KV DB subsystem

The framework supports KV DBs in form of Redis.

The way we work with KV DB connections differs from the SQL DB subsystem. This time, we should specify not the connection instance, but rather its name. Certainly, the connection should be pre-created, otherwise you'll get the error KVDBConnectionDoesNotExist.

myFlow :: T.KVDBKey -> L.Flow (T.TxResult (Maybe T.KVDBValue))
myFlow k = L.runKVDB "redis" $ L.multiExec $ do
  L.setTx k "bbb"
  res <- L.getTx k
  L.delTx [k]
  pure res

-- returns T.TxSuccess (Just "bbb")

See tests for more info:

• KVDBSpec
• KVDBArtSpec

Concurrency

EulerHS is a concurrent framework. It can be easily used with Servant or Scotty for building web services and RESTful applications. The runtime of the framework is concurrent and is capable to handle parallel flows. Although the Runtime is thread-safe, it doesn't free you from writing thread safe flows. The framework exposes a set of operations which may produce race conditions if used wrongly and/or from different threads:

getOption setOption delOption

initSqlDBConnection deinitSqlDBConnection getSqlDBConnection

Forking and awaiting child flows

It's possible to fork controllable flows and await for their results. This subsystem can be used to compose async-like flows, but the main case is parallel execution.

Flow has the following methods to work with forked flows:

forkFlow  :: Description -> Flow () -> Flow ()
forkFlow' :: Description -> Flow a -> Flow (Awaitable (Either Text a))
await :: Maybe Microseconds -> Awaitable (Either Text a) -> m (Either AwaitingError a)

Notice that in the current framework there are no methods for killing a forked flow.

myFlow :: L.Flow (Maybe String, Maybe String)
myFlow = do
  awaitable1 <- L.forkFlow' "Child Flow 1" (L.runIO (threadDelay 10000) >> pure "1")
  awaitable2 <- L.forkFlow' "Child Flow 2" (L.runIO (threadDelay 100000) >> pure "2")
  mbRes1 <- L.await Nothing awaitable1
  mbRes2 <- L.await Nothing awaitable2
  pure (mbRes1, mbRes2)

  -- Returns (Just "1", Just "2") after approximately 100000 ms.

Building the framework

See BUILD.md

Testing the framework

You can run stack test to see if your system is ready, and the framework can be used.

Integration tests for SQL backends

There are disabled tests for MySQL and Postgres DB backends. To run them:

• Create DB tables and fill with the appropriate data using these sql queries:
  • MySQLDBSpec.sql
  • PostgresDBSpec.sql
• Uncomment the corresponding specs in lib/euler-hs/testSqlDB/Main.hs
• Run tests: stack test euler-hs:sql

EulerHS tutorials and template projects

• Tutorial
• Beam query examples
• Echo server demo project

Background materials

• Hierarchical Free Monads
• Automatic whitebox testing with Free monads