persistent-typed-db: Type safe access to multiple database schemata.

[ bsd3, library, web ] [ Propose Tags ]

See for more details, examples, and fun.

[Skip to Readme]
Versions [faq],,
Change log
Dependencies aeson, base (>=4.7 && <5), bytestring, conduit (>=1.3.0), http-api-data, monad-logger, path-pieces, persistent (>=2.10.0), persistent-template (>=2.7.0), resource-pool, resourcet (>=1.2.0), template-haskell, text, transformers [details]
License BSD-3-Clause
Copyright 2017 Matt Parsons
Author Matt Parsons
Category Web
Home page
Bug tracker
Source repo head: git clone
Uploaded by parsonsmatt at Wed May 1 15:41:33 UTC 2019
Distributions NixOS:, Stackage:
Downloads 150 total (85 in the last 30 days)
Rating 2.0 (votes: 1) [estimated by rule of succession]
Your Rating
  • λ
  • λ
  • λ
Status Hackage Matrix CI
Docs available [build log]
Last success reported on 2019-05-01 [all 1 reports]


[Index] [Quick Jump]


Maintainer's Corner

For package maintainers and hackage trustees

Readme for persistent-typed-db-

[back to package description]


Build Status

This library defines an alternate SqlBackend type for the Haskell persistent database library. The type has a phantom type parameter which allows you to write queries against multiple databases safely.

The Problem

The persistent library uses a "handle pattern" with the SqlBackend type. The SqlBackend (or Pool SqlBackend) is used to provide access to "The Database." To access the database, you generally use a function like:

runDB :: SqlPersistT m a -> App a

However, you may have two (or more!) databases. You might have multiple database "runner" functions, like this:

    :: SqlPersistT m a -> App a
    :: SqlPersistT m a -> App a

Unfortunately, this isn't safe. The schemas differ, and there's nothing preventing you from using the wrong runner for the query at hand.

This library allows you to differentiate between database schemata. To demonstrate usage, we'll start with a pair of ordinary persistent quasiquoter blocks:

share [ mkPersist sqlSettings, mkMigrate "migrateAll" ] [persistLowerCase|

    name Text
    age  Int

    deriving Show Eq

share [ mkPersist sqlSettings, mkMigrate "migrateAll" ] [persistLowerCase|

    createdAt UTCTime
    reason    Text

    deriving Show Eq

These two definitions correspond to different databases. The persistent library uses the SqlPersistT m monad transformer for queries. This type is a synonym for ReaderT SqlBackend m. Many of the functions defined in persistent have a signature like this:

get :: (MonadIO m, PersistEntityBackend record ~ backend)
    => Key record
    -> ReaderT backend m (Maybe record)

It requires that the entity is compatible with the query monad. We're going to substitute User for the type variable record. In the initial schema definition, the PersistEntityBackend User is defined as SqlBackend. So the type of get, in the original definition, is this:

get :: (MonadIO m, PersistEntityBackend User ~ SqlBackend)
    => Key record
    -> ReaderT SqlBackend m (Maybe User)

If we look at the type of get specialized to AuxRecord, we see this:

get :: (MonadIO m, PersistEntityBackend AuxRecord ~ SqlBackend)
    => Key record
    -> ReaderT SqlBackend m (Maybe AuxRecord)

This means that we might be able to write a query like this:

    :: MonadIO m
    => SqlPersistT m (Maybe User, Maybe AuxRecord)
impossibleQuery = do
  muser <- get (UserKey 1)
  maux <- get (AuxRecordKey 1)
  pure (muser, maux)

This query will fail at runtime, since the entities exist on different schemata. Likewise, there's nothing in the types to stop you from running a query against the wrong backend:

app = do
    runMainDB $ get (AuxRecordKey 3)
    runAuxDb $ get (UserKey 3)

The Solution

Let's solve this problem.

Declaring the Schema

We are going to create an empty datatype tag for each schema, and then we're going to use mkSqlSettingsFor instead of sqlSettings.

data MainDb
data AuxDb

share [ mkPersist (mkSqlSettingsFor ''MainDb), mkMigrate "migrateAll" ] [persistLowerCase|

    name Text
    age  Int

    deriving Show Eq

share [ mkPersist (mkSqlSettingsFor ''AuxDb), mkMigrate "migrateAll" ] [persistLowerCase|

    createdAt UTCTime
    reason    Text

    deriving Show Eq

This changes the type of the PersistEntityBackend record for each entity defined in the QuasiQuoter. The previous type of PersistEntityBackend User was SqlBackend, but with this change, it is now SqlFor MainDb. Likewise, the type of PersistEntityBackend AuxRecord has become SqlFor AuxDb.

Using the Schema

Let's look at the new type of get for these two records:

get :: (MonadIO m, PersistEntityBackend User ~ SqlFor MainDb)
    => Key record
    -> ReaderT (SqlFor MainDb) m (Maybe User)

get :: (MonadIO m, PersistEntityBackend AuxRecord ~ SqlFor AuxDb)
    => Key record
    -> ReaderT (SqlFor AuxDb) m (Maybe AuxRecord)

Now that the monad type is different, we can't use them in the same query. Our previous impossibleQuery now fails with a type error.

The persistent-typed-db library defines a type synonym for ReaderT. It is similar to the SqlPersistT synonym:

type SqlPersistT       = ReaderT SqlBackend
type SqlPersistTFor db = ReaderT (SqlFor db)

When using this library, it is a good idea to define a type snynonym for your databases as well. So we might also write:

type MainDbT = SqlPersistTFor MainDb
type AuxDbT  = SqlPersistTFor AuxDb

The type of our runner functions has changed, as well. Before, we accepted a SqlPersistT, but now, we'll accept the right query type for the database:

runMainDb :: MainDbT m a -> App a

runAuxDb  :: AuxDbT m a -> App a

We'll cover how to define these runner functions soon.

Defining the Runner Function

persistent defines a function runSqlPool that is useful for running a SQL action. The type is essentially this:

    :: (MonadUnliftIO m, IsSqlBackend backend)
    => ReaderT backend m a
    -> Pool backend
    -> m a

persistent-typed-db defines a function that is a drop in replacement for this, called runSqlPoolFor.

    :: (MonadUnliftIO m)
    => SqlPersistTFor db m a
    -> ConnectionPoolFor db
    -> m a

It is defined by generalizing the input query and pool, and delegating to runSqlPool.

runSqlPoolFor query conn =
    runSqlPool (generalizeQuery query) (generalizePool conn)

Sometimes, you'll have some function that is in SqlPersistT that you want to use on a specialized database. This can occur with raw queries, like rawSql and friends, or other queries/actions that are not tied to a PersistEntityBackend type. In this case, you'll want to use specializeQuery. You will likely want to define type-specified helpers that are aliases for specializeQuery:

toMainQuery :: SqlPersistT m a -> MainDbT m a
toMainQuery = specializeQuery

toAuxQuery :: SqlPersistT m a -> AuxDbT m a
toAuxQuery = specializeQuery

Constructing the Pools

persistent (and the relevant database-specific libraries) define many functions for creating connections. We'll use createPostgresqlPool as an example. This is one place where you do need to be careful, as you are tagging the database pool with the database type.

To create a specific database pool, we'll map specializePool over the result of createPostgresqlPool:

    :: ConnectionString
    -> Int
    -> IO (ConnectionPoolFor db)
createPostgresqlPoolFor connStr i =
    specializePool <$> createPostgresqlPool connStr i

It is a good idea to make specialized variants of this function to improve type inference and errors:

createMainPool :: ConnectionString -> Int -> IO (ConnectionPoolFor MainDb)
createMainPool = createPostgresqlPoolFor

createAuxPool  :: ConnectionString -> Int -> IO (ConnectionPoolFor AuxDb)
createAuxPool  = createPostgresqlPoolFor

It is common to use with-style functions with these, as well. These functions automate closure of the database resources. We can specialize these functions similarly:

    :: forall db m a
     .  (MonadLogger m, MonadUnliftIO m)
    => ConnectionString
    -> Int
    -> (ConnectionPoolFor db -> m a)
    -> m a
withPoolFor connStr conns action =
    withPostgresqlPool connStr conns $ \genericPool ->
        action (specializePool genericPool)