= Database Transactions 

Sequel uses autocommit mode by default for all of its database adapters, so in general in Sequel if you want to use database transactions, you need to be explicit about it.  There are a few cases where transactions are used implicitly by default:

* Dataset#import to insert many records at once
* Model#save
* Model#destroy
* Migrations if the database supports transactional schema
* A few model plugins

Everywhere else, it is up to you to use a database transaction if you want to.

== Basic Transaction Usage

In Sequel, the <tt>Database#transaction</tt> method should be called if you want to use a database transaction.  This method must be called with a block.  If the block does not raise an exception, the transaction is committed:

  DB.transaction do # BEGIN
    DB[:foo].insert(1) # INSERT
  end # COMMIT
   
If the block raises a Sequel::Rollback exception, the transaction is rolled back, but no exception is raised outside the block:

  DB.transaction do # BEGIN
    raise Sequel::Rollback
  end # ROLLBACK
  # no exception raised

If any other exception is raised, the transaction is rolled back, and the exception is raised outside the block:

  DB.transaction do # BEGIN
    raise ArgumentError
  end # ROLLBACK
  # ArgumentError raised

If you want Sequel::Rollback exceptions to be reraised, use the <tt>:rollback => :reraise</tt> option:

  DB.transaction(:rollback => :reraise) do # BEGIN
    raise Sequel::Rollback
  end # ROLLBACK
  # Sequel::Rollback raised

If you always want to rollback (useful for testing), use the <tt>:rollback => :always</tt> option:

  DB.transaction(:rollback => :always) do # BEGIN
    DB[:foo].insert(1) # INSERT
  end # ROLLBACK
  # no exception raised

If you want to check whether you are currently in a transaction, use the Database#in_transaction? method:

  DB.in_transaction? # false
  DB.transaction do
    DB.in_transaction? # true
  end

== Transaction Hooks

You can add hooks to an in progress transaction that are called after the transaction commits or rolls back:

  x = nil
  DB.transaction do
    DB.after_commit{x = 1}
    DB.after_rollback{x = 2}
    x # nil
  end
  x # 1
  
  x = nil
  DB.transaction do
    DB.after_commit{x = 1}
    DB.after_rollback{x = 2}
    raise Sequel::Rollback
  end
  x # 2

== Nested Transaction Calls / Savepoints

You can nest calls to transaction, which by default just reuses the existing transaction:

  DB.transaction do # BEGIN
    DB.transaction do
      DB[:foo].insert(1) # INSERT
    end
  end # COMMIT

You can use the <tt>:savepoint => true</tt> option in the inner transaction to explicitly use a savepoint (if the database supports it):

  DB.transaction do # BEGIN
    DB.transaction(:savepoint => true) do # SAVEPOINT
      DB[:foo].insert(1) # INSERT
    end # RELEASE SAVEPOINT
  end # COMMIT

If a Sequel::Rollback exception is raised inside the savepoint block, it will only rollback to the savepoint:

  DB.transaction do # BEGIN
    DB.transaction(:savepoint => true) do # SAVEPOINT
      raise Sequel::Rollback
    end # ROLLBACK TO SAVEPOINT
    # no exception raised
  end # COMMIT

Other exceptions, unless rescued inside the outer transaction block, will rollback the savepoint and the outer transactions, since they are reraised by the transaction code:

  DB.transaction do # BEGIN
    DB.transaction(:savepoint => true) do # SAVEPOINT
      raise ArgumentError
    end # ROLLBACK TO SAVEPOINT
  end # ROLLBACK
  # ArgumentError raised

== Prepared Transactions / Two-Phase Commit

Sequel supports database prepared transactions on PostgreSQL, MySQL, and H2.  With prepared transactions, at the end of the transaction, the transaction is not immediately committed (it acts like a rollback).  Later, you can call +commit_prepared_transaction+ to commit the transaction or +rollback_prepared_transaction+ to roll the transaction back.  Prepared transactions are usually used with distributed databases to make sure all databases commit the same transaction or none of them do.

To use prepared transactions in Sequel, you provide a string as the value of the :prepare option:

  DB.transaction(:prepare => 'foo') do # BEGIN
    DB[:foo].insert(1) # INSERT
  end # PREPARE TRANSACTION 'foo'
   
Later, you can commit the prepared transaction:

  DB.commit_prepared_transaction('foo')

or roll the prepared transaction back:

  DB.rollback_prepared_transaction('foo')

== Transaction Isolation Levels

The SQL standard supports 4 isolation levels: READ UNCOMMITTED, READ COMMITTED, REPEATABLE READ, and SERIALIZABLE.  Not all databases implement the levels as specified in the standard (or implement the levels at all), but on most databases, you can specify which transaction isolation level you want to use via the :isolation option to <tt>Database#transaction</tt>.  The isolation level is specified as one of the following symbols: :uncommitted, :committed, :repeatable, and :serializable.  Using this option make Sequel use the correct transaction isolation syntax for your database:

  DB.transaction(:isolation => :serializable) do # BEGIN
    # SET TRANSACTION ISOLATION LEVEL SERIALIZABLE
    DB[:foo].insert(1) # INSERT
  end # COMMIT

== Automatically Restarting Transactions

Sequel offers the ability to automatically restart transactions if specific types of errors are detected.  For example, if you want to automatically restart a transaction if a serialization failure is detected:

  DB.transaction(:isolation => :serializable, :retry_on=>[Sequel::SerializationFailure]) do
    ModelClass.find_or_create(:name=>'Foo')
  end

At the serializable transaction isolation level, find_or_create may raises a Sequel::SerializationFailure exception if multiple threads simultaneously run that code.  With the :retry_on option set, the transaction will be automatically retried until it succeeds.

Note that automatic retrying should not be used unless the entire transaction
block is idempotent, as otherwise it can cause non-idempotent
behavior to execute multiple times.  For example, with the following code:

  DB.transaction(:isolation => :serializable, :retry_on=>[Sequel::SerializationFailure]) do
    logger.info 'Ensuring existence of ModelClass with name Foo'
    ModelClass.find_or_create(:name=>'Foo')
  end

The logger.info method will be called multiple times if there is a serialization failure.

The :num_retries option can be used to set the maxmimum number of times to retry.  It is set to 5 times by default.