8. Error Handling

In this chapter we examine a set of combinators that allow us to construct programs that trap and handle exceptions.

Setting Up

import doobie.imports._
import scalaz._, Scalaz._
import scalaz.concurrent.Task

val xa = DriverManagerTransactor[Task](
  "org.postgresql.Driver", "jdbc:postgresql:world", "postgres", ""
)

import xa.yolo._

About Exceptions

Exceptions are a fact of life when interacting with databases, and they are largely nondeterministic; whether an operation will succeed or not depends on unpredictable factors like network health, the current contents of tables, locking state, and so on. So we must decide whether to compute everything in a disjunction like EitherT[ConnectionIO, Throwable, A] or allow exceptions to propagate until they are caught explicitly. doobie adopts the second strategy: exceptions are allowed to propagate and escape unless handled explicitly (exactly as IO and Task work). This means when a doobie action (transformed to some target monad) is executed, exceptions can escape.

There are three main types of exceptions that are likely to arise:

1. Various types of IOException can happen with any kind of I/O, and these exceptions tend to be unrecoverable.
2. Database exceptions, typically as a generic SQLException with a vendor-specific SQLState identifying the specific error, are raised for common situations such as key violations. Some vendors (PostgreSQL for instance) publish a table of error codes, and in these cases doobie can provide a matching set of exception-handling combinators. However in most cases the error codes must be passed down as folklore or discovered by experimentation. There exist the XOPEN and SQL:2003 standards, but it seems that no vendor adheres closely to these specifications. Some of these errors are recoverable and others aren’t.
3. doobie will raise an InvariantViolation in response to invalid type mappings, unknown JDBC constants returned by drivers, observed NULL values, and other violations of invariants that doobie assumes. These exceptions indicate programmer error or driver non-compliance and are generally unrecoverable.

The Catchable Typeclass and Derived Combinators

All doobie monads have associated instances of the scalaz.Catchable typeclass, and the provided interpreter requires all target monads to have an instance as well. Catchable provides two operations:

• attempt converts M[A] into M[Throwable \/ A]
• fail constructs an M[A] that fails with a provided Throwable

So any doobie program can be lifted into a disjunction simply by adding .attempt.

scala> val p = 42.point[ConnectionIO]
p: doobie.imports.ConnectionIO[Int] = Return(42)

scala> p.attempt
res2: doobie.imports.ConnectionIO[scalaz.\/[Throwable,Int]] = Suspend(scalaz.Coyoneda$$anon$22@133a171b)

From the .attempt combinator we derive the following, available as combinators and as syntax:

• attemptSome allows you to catch only specified Throwables.
• except recovers with a new action.
• exceptSome same, but only for specified Throwables.
• onException executes an action on failure, discarding its result.
• ensuring executes an action in all cases, generalizing finally.

From these we can derive combinators that only pay attention to SQLException:

• attemptSql is like attempt but only traps SQLException.
• attemptSomeSql traps only specified SQLExceptions.
• exceptSql recovers from a SQLException with a new action.
• onSqlException executes an action on SQLException and discards its result.

And finally we have a set of combinators that focus on SQLStates.

• attemptSqlState is like attemptSql but yields M[SQLState \/ A].
• attemptSomeSqlState traps only specified SQLStates.
• exceptSqlState recovers from a SQLState with a new action.
• exceptSomeSqlState recovers from specified SQLStates with a new action.

See the ScalaDoc for more information.

Example: Unique Constraint Violation

Ok let’s set up a person table again, using a slightly different formulation just for fun. Note that the name column is marked as being unique.

scala> List(sql"""DROP TABLE IF EXISTS person""",
     |      sql"""CREATE TABLE person (
     |              id    SERIAL,
     |              name  VARCHAR NOT NULL UNIQUE
     |            )""").traverse(_.update.quick).void.run
  0 row(s) updated
  0 row(s) updated

Alright, let’s define a Person data type and a way to insert instances.

case class Person(id: Int, name: String)

def insert(s: String): ConnectionIO[Person] = {
  sql"insert into person (name) values ($s)"
    .update.withUniqueGeneratedKeys("id", "name")
}

The first insert will work.

scala> insert("bob").quick.run
  Person(1,bob)

The second will fail with a unique constraint violation.

scala> try {
     |   insert("bob").quick.run
     | } catch {
     |   case e: java.sql.SQLException => 
     |     println(e.getMessage)
     |     println(e.getSQLState)
     | }
ERROR: duplicate key value violates unique constraint "person_name_key"
  Detail: Key (name)=(bob) already exists.
23505

So let’s change our method to return a String \/ Person by using the attemptSomeSql combinator. This allows us to specify the SQLState value that we want to trap. In this case the culprit "23505" (yes, it’s a string) is provided as a constant in the contrib-postgresql add-on.

import doobie.contrib.postgresql.sqlstate.class23.UNIQUE_VIOLATION

def safeInsert(s: String): ConnectionIO[String \/ Person] =
  insert(s).attemptSomeSqlState {
    case UNIQUE_VIOLATION => "Oops!"
  }

Given this definition we can safely attempt to insert duplicate records and get a helpful error message rather than an exception.

scala> safeInsert("bob").quick.run
  -\/(Oops!)

scala> safeInsert("steve").quick.run
  \/-(Person(4,steve))