Skip to content

Part 1

In this tutorial sequence, we'll walk through creating a schema for a simple shopping cart database. We'll start by defining a user table. Then, we'll show how beam makes it easy to manipulate data in our database. Finally, we'll demonstrate how beam lets us declare type-safe and composable queries.

Beam Module Structure

Beam makes extensive use of GHC's Generics mechanism. This extension means beam does not need to rely on template haskell.

To start defining beam schemas and queries, you only need to import the Database.Beam module. To interface with an actual database, you'll need to import one of the database backends. We'll see how to use the Sqlite backend here (found in the beam-sqlite package). Now, open up a GHCi prompt for us to use. Make sure to get the beam-core and beam-sqlite packages.

$ stack repl --package beam-core --package beam-sqlite --package sqlite-simple --package beam-migrate

This will put you into a GHCi prompt with the beam-core and beam-sqlite packages available. We also include the sqlite-simple package. Beam mainly manages querying and data marshalling. Connections to the backends are done via backend specific packages. In this case, beam-sqlite uses the sqlite-simple backend.

Before starting, we'll need to enable some extensions.

> :set -XDeriveGeneric -XGADTs -XOverloadedStrings -XFlexibleContexts -XFlexibleInstances -XTypeFamilies -XTypeApplications

And import some modules...

import Database.Beam
import Database.Beam.Sqlite

import Data.Text (Text)

Defining our first table

Beam tables are regular Haskell data types with a bit of scaffolding. Thankfully, the magic of the modern Haskell type system allows us to remove the overhead and the syntactic fuzz of the scaffolding in most situations.

We start by declaring a data structure named UserT. As a matter of convention, Beam table types are suffixed with 'T'. Table types have only one constructor. Again, as a matter of convention, the constructor has the same name as the table, but without the 'T' suffix. We'll soon see the reason for this convention.

In this tutorial, I'll prefix all record selectors with an underscore. This is a matter of personal preference. One reason for the prefix is that it plays nicely with the lens library. Beam does not necessitate the use of lens (in fact Beam includes its own mechanism to generically derive van Laarhoven lenses), but I recognize that some programmers use lens quite a lot.

data UserT f
    = User
    { _userEmail     :: Columnar f Text
    , _userFirstName :: Columnar f Text
    , _userLastName  :: Columnar f Text
    , _userPassword  :: Columnar f Text }
    deriving Generic

This data type might look very complicated, so I'd like to show you that it's not that scary. Let's see if we can use GHCi to help us.

Prelude Database.Beam.Sqlite Database.Beam Data.Text> :t User
  :: Columnar f Text
     -> Columnar f Text -> Columnar f Text -> Columnar f Text -> UserT f

Hmm... That did not help much. However, consider the type of the following:

Prelude Database.Beam.Sqlite Database.Beam Data.Text> :t (\email firstName lastName password -> User email firstName lastName password :: UserT Identity)
(\email firstName lastName password -> User email firstName lastName password :: UserT Identity)
  :: Text -> Text -> Text -> Text -> UserT Identity

Woah! That looks a lot like what we'd expect if we had declared the type in the "regular" Haskell way:

data User = User
          { _userEmail     :: Text
          , _userFirstName :: Text
          , _userLastName  :: Text
          , _userPassword  :: Text }

This functionality is due to the fact that Columnar is a type family defined such that for any x, Columnar Identity x = x. This strategy is known as defunctionalization 1.

Knowing this, let's define a type synonym to make our life easier.

type User = UserT Identity
type UserId = PrimaryKey UserT Identity

Now you can see why we named the type of the table UserT and its constructor User. This allows us to use the "regular" User constructor to construct values of type User. We can use the StandaloneDeriving and TypeSynonymInstances extensions to derive instances of Show and Eq for the 'regular' datatype.

> :set -XStandaloneDeriving -XTypeSynonymInstances

Now we can derive Show and Eq instances.

deriving instance Show User
deriving instance Eq User

Note that this does require us to use an explicit type signature where we otherwise wouldn't. For example,

Prelude Database.Beam.Sqlite Database.Beam Data.Text> User "" "John" "Smith" "password!"

<interactive>:46:2: error:
    * No instance for (Show (UserT f0)) arising from a use of &lsquo;print&rsquo;
    * In a stmt of an interactive GHCi command: print it

Here, GHC is complaining that it cannot infer the type of the f parameter based on the values we've supplied. This is because the Columnar type family is non-injective. However, an explicit type annotation fixes it all up.

Prelude Database.Beam.Sqlite Database.Beam Data.Text> User "" "John" "Smith" "password!" :: User
User {_userEmail = "", _userFirstName = "John", _userLastName = "Smith", _userPassword = "password!"}

Usually, you won't need to deal with this, as you'll explicitly annotate your top-level functions to use the User type.

Teaching Beam about our table

We've defined a type that can represent the data in our table. Now, let's inform beam that we'd like to use UserT as a table.

All beam tables need to implement the Beamable type class. Due to GHC's DeriveGeneric and DefaultSignatures extensions, all these methods can be written for us by the compiler at compile-time!

instance Beamable UserT

Additionally, all beam tables must implement the Table type class, which we can use to declare a primary key.

The only thing we need to provide is the type of the primary keys for users, and a function that can extract the primary key from any UserT f object. To do this, add the following lines to the instance declaration.

instance Table UserT where
    data PrimaryKey UserT f = UserId (Columnar f Text) deriving Generic
    primaryKey = UserId . _userEmail
instance Beamable (PrimaryKey UserT)

The data declaration is similar to a toplevel data definition, construct a key for UserT with the UserId constructor like a regular table.

userKey = UserId ""


The standalone Beamable instances are quite ugly. Luckily, the new deriving strategies extension in GHC 8.2 will allow us to write the Beamable instance 'in-line', so we can write deriving (Generic, Beamable) instead.

Defining our database

Now that we have our table, we're going to define a type to hold information about our database. Defining our database is going to follow the same pattern as defining a table. We'll define a higher-kinded datatype and then declare an instance of Database, and let the compiler figure most of it out.

Tables are a collection of Columnar values. Databases are a collection of entities, such as tables. Many database systems can also hold other entities (such as views, domain types, etc). Beam allows you to declare these as well 2.

Our database consists of only one table.

data ShoppingCartDb f = ShoppingCartDb
                      { _shoppingCartUsers :: f (TableEntity UserT) }
                        deriving Generic

instance Database be ShoppingCartDb

The next step is to create a description of the particular database we'd like to create. This involves giving each of the tables in our database a name. If you've named all your database selectors using camel case, beam can automatically figure out what all the table names should be. If you haven't, or you have multiple tables holding the same type in your database, you might have to manually name your tables. For now, we'll let beam do the hard work 3.

shoppingCartDb :: DatabaseSettings be ShoppingCartDb
shoppingCartDb = defaultDbSettings

Adding users to our database

Let's add some users to our database. As we said above, beam is backend-agnostic. However, backend integration libraries are maintained in the official beam repository. The beam-sqlite package offers straightforwards integration with the sqlite-simple library.

First, let's create a sqlite3 database with the right schema. Open up terminal, and do

$ sqlite3 shoppingcart1.db
SQLite version 3.14.0 2016-07-26 15:17:14
Enter ".help" for usage hints.
sqlite> CREATE TABLE cart_users (email VARCHAR NOT NULL, first_name VARCHAR NOT NULL, last_name VARCHAR NOT NULL, password VARCHAR NOT NULL, PRIMARY KEY( email ));

Now, let's open the database in Haskell.

import Database.SQLite.Simple

conn <- open "shoppingcart1.db"


Previous versions of beam would attempt automatic schema migration. This is dangerous and not required for many use cases. A more powerful implementation of this functionality has been moved into the optional beam-migrate package. See the appropriate documentation

Now let's add a few users. We'll give each user an MD5 encoded password too. We'll use the runBeamSqliteDebug function (supplied by beam-sqlite) to output the statements that beam would normally run. In production, you'd use the runBeamSqlite function, or use the backend integration packages to directly use the underlying backend library.

runBeamSqliteDebug putStrLn {- for debug output -} conn $ runInsert $
insert (_shoppingCartUsers shoppingCartDb) $
insertValues [ User "" "James" "Smith" "b4cc344d25a2efe540adbf2678e2304c" {- james -}
             , User "" "Betty" "Jones" "82b054bd83ffad9b6cf8bdb98ce3cc2f" {- betty -}
             , User "" "Sam" "Taylor" "332532dcfaa1cbf61e2a266bd723612c" {- sam -} ]

The runInsert function runs an insert statement, which we construct using the insert function. Since we're inserting concrete values, we use the insertValues function to supply the values. We can also use the insertExpressions function to insert arbitrary SQL expressions, or the insertFrom to insert the results of an arbitrary select (the INSERT INTO .. SELECT .. syntax).

Because we're in debug mode, we'll see the SQL that beam is running:

INSERT INTO "cart_users"("email", "first_name", "last_name", "password") VALUES (?, ?, ?, ?), (?, ?, ?, ?), (?, ?, ?, ?)
-- With values: [SQLText "",SQLText "James",SQLText "Smith",SQLText "b4cc344d25a2efe540adbf2678e2304c",SQLText "",SQLText "Betty",SQLText "Jones",SQLText "82b054bd83ffad9b6cf8bdb98ce3cc2f",SQLText "",SQLText "Sam",SQLText "Taylor",SQLText "332532dcfaa1cbf61e2a266bd723612c"]

The ? represent the values passed to the database (beam uses the backend's value interpolation to avoid SQL injection attacks).

Querying the database

Now let's write some queries for the database. Let's get all the users we just added. Click between the tabs to see the SQL and console output generated

let allUsers = all_ (_shoppingCartUsers shoppingCartDb)

runBeamSqliteDebug putStrLn conn $ do
  users <- runSelectReturningList $ select allUsers
  mapM_ (liftIO . putStrLn . show) users
SELECT "t0"."email" AS "res0",
       "t0"."first_name" AS "res1",
       "t0"."last_name" AS "res2",
       "t0"."password" AS "res3"
FROM "cart_users" AS "t0";

-- With values: []
User {_userEmail = "", _userFirstName = "James", _userLastName = "Smith", _userPassword = "b4cc344d25a2efe540adbf2678e2304c"}
User {_userEmail = "", _userFirstName = "Betty", _userLastName = "Jones", _userPassword = "82b054bd83ffad9b6cf8bdb98ce3cc2f"}
User {_userEmail = "", _userFirstName = "Sam", _userLastName = "Taylor", _userPassword = "332532dcfaa1cbf61e2a266bd723612c"}


The -- at the ends of the console output lines are an artifact of the documentation build process. They won't appear in your console.

Next let's suppose you wanted to sort the users into order by their first name, and then descending by their last name. We can use the orderBy_ function to order the query results. This is similar to the sortBy function for lists.

let sortUsersByFirstName = orderBy_ (\u -> (asc_ (_userFirstName u), desc_ (_userLastName u))) (all_ (_shoppingCartUsers shoppingCartDb))

runBeamSqliteDebug putStrLn conn $ do
  users <- runSelectReturningList $ select sortUsersByFirstName
  mapM_ (liftIO . putStrLn . show) users
SELECT "t0"."email" AS "res0",
       "t0"."first_name" AS "res1",
       "t0"."last_name" AS "res2",
       "t0"."password" AS "res3"
FROM "cart_users" AS "t0"
ORDER BY "t0"."first_name" ASC,
         "t0"."last_name" DESC;

-- With values: []
User {_userEmail = "", _userFirstName = "Betty", _userLastName = "Jones", _userPassword = "82b054bd83ffad9b6cf8bdb98ce3cc2f"}
User {_userEmail = "", _userFirstName = "James", _userLastName = "Smith", _userPassword = "b4cc344d25a2efe540adbf2678e2304c"}
User {_userEmail = "", _userFirstName = "Sam", _userLastName = "Taylor", _userPassword = "332532dcfaa1cbf61e2a266bd723612c"}

We can use limit_ and offset_ in a similar manner to take and drop respectively.

let boundedQuery :: Q SqliteSelectSyntax _ _ _
    boundedQuery = limit_ 1 $ offset_ 1 $
                   orderBy_ (asc_ . _userFirstName) $
                   all_ (_shoppingCartUsers shoppingCartDb)

runBeamSqliteDebug putStrLn conn $ do
  users <- runSelectReturningList (select boundedQuery :: SqlSelect SqliteSelectSyntax _)
  mapM_ (liftIO . putStrLn . show) users
SELECT "t0"."email" AS "res0",
       "t0"."first_name" AS "res1",
       "t0"."last_name" AS "res2",
       "t0"."password" AS "res3"
FROM "cart_users" AS "t0"
ORDER BY "t0"."first_name" ASC

-- With values: []
User {_userEmail = "", _userFirstName = "James", _userLastName = "Smith", _userPassword = "b4cc344d25a2efe540adbf2678e2304c"}


The type signatures above are not necessary when run in GHCi. The documentation automates building and testing the queries, but the above results in strange type errors in GHC. This may be a compiler type-inference bug. More investigation is being carried out.

Nevertheless, this shows how you could limit your queries to only work in a particular syntax.

The _ are type holes, which means GHC will happily infer these types, if the PartialTypeSignatures extension is turned on.


Sometimes we also want to group our data together and perform calculations over the groups of data. SQL calls these aggregations.

The simplest aggregation is counting. We use the aggregate_ function to create aggregations. For example, to count all users, we can use the countAll_ aggregation. We also use the runSelectReturningOne function to get at most one record from the database.

let userCount = aggregate_ (\u -> as_ @Int countAll_) (all_ (_shoppingCartUsers shoppingCartDb))

runBeamSqliteDebug putStrLn conn $ do
  Just c <- runSelectReturningOne $ select userCount
  liftIO $ putStrLn ("We have " ++ show c ++ " users in the database")
FROM "cart_users" AS "t0";

-- With values: []
We have 3 users in the database


countAll_ is happy to unmarshal into any Integral type, so we use as_ to constrain the type to Int.

Maybe we'd like something a little more interesting, such as the number of users for each unique first name. We can also express these aggregations using the aggregate_ function. In order to get interesting results, we'll need to add more users to our database. We'll demonstrate using withDatabase to silence the debug messages.

runBeamSqliteDebug putStrLn conn $
  runInsert $
  insert (_shoppingCartUsers shoppingCartDb) $
  insertValues [ User "" "James" "Pallo" "b4cc344d25a2efe540adbf2678e2304c" {- james -}
               , User "" "Betty" "Sims" "82b054bd83ffad9b6cf8bdb98ce3cc2f" {- betty -}
               , User "" "James" "O'Reily" "b4cc344d25a2efe540adbf2678e2304c" {- james -}
               , User "" "Sam" "Sophitz" "332532dcfaa1cbf61e2a266bd723612c" {- sam -}
               , User "" "Sam" "Jely" "332532dcfaa1cbf61e2a266bd723612c" {- sam -} ]
INSERT INTO "cart_users"("email",
        ?), (?,
             ?), (?,
                  ?), (?,
                       ?), (?,

-- With values: [SQLText "",SQLText "James",SQLText "Pallo",SQLText "b4cc344d25a2efe540adbf2678e2304c",SQLText "",SQLText "Betty",SQLText "Sims",SQLText "82b054bd83ffad9b6cf8bdb98ce3cc2f",SQLText "",SQLText "James",SQLText "O'Reily",SQLText "b4cc344d25a2efe540adbf2678e2304c",SQLText "",SQLText "Sam",SQLText "Sophitz",SQLText "332532dcfaa1cbf61e2a266bd723612c",SQLText "",SQLText "Sam",SQLText "Jely",SQLText "332532dcfaa1cbf61e2a266bd723612c"]

Now we can use aggregate_ to both group by a user's first name, and then count the number of users.

let numberOfUsersByName = aggregate_ (\u -> (group_ (_userFirstName u), as_ @Int countAll_)) $
                          all_ (_shoppingCartUsers shoppingCartDb)

runBeamSqliteDebug putStrLn conn $ do
  countedByName <- runSelectReturningList $ select numberOfUsersByName
  mapM_ (liftIO . putStrLn . show) countedByName
SELECT "t0"."first_name" AS "res0",
       COUNT(*) AS "res1"
FROM "cart_users" AS "t0"
GROUP BY "t0"."first_name";

-- With values: []


In this tutorial, we've covered creating a database schema, opening up a beam database, inserting values into the database, and querying values from them. We used the knowledge we learned to create a partial shopping cart database that contains information about users. In the next tutorial, we'll delve deeper into the some of the query types and show how we can create relations between tables. We'll also use the monadic query interface to create SQL joins.

Until next time!

If you have any questions about beam, feel free to send them to . Pull requests and bug reports are welcome on GitHub.

  1. Thanks to various bloggers for pointing this out. You can read more about this technique here

  2. Adding entities other than tables is covered in more depth in the user guide

  3. More on the default naming conventions can be found in the models section of the user guide. We'll talk about how to override defaults in the next sections.