Transactions are protective blocks where SQL statements are only permanent if they can all succeed as one atomic action. The classic example is a transfer between two accounts where you can only have a deposit if the withdrawal succeeded and vice versa. Transactions enforce the integrity of the database and guard the data against program errors or database break-downs. So basically you should use transaction blocks whenever you have a number of statements that must be executed together or not at all. Example:
ActiveRecord::Base.transaction do david.withdrawal(100) mary.deposit(100) end
This example will only take money from David and give to Mary if neither withdrawal nor deposit raises an exception. Exceptions will force a ROLLBACK that returns the database to the state before the transaction was begun. Be aware, though, that the objects will not have their instance data returned to their pre-transactional state.
Different Active Record classes in a single transaction
Though the transaction class method is called on some Active Record class, the objects within the transaction block need not all be instances of that class. This is because transactions are per-database connection, not per-model.
In this example a Balance record is transactionally saved even though transaction is called on the Account class:
Account.transaction do balance.save! account.save! end
Note that the transaction method is also available as a model instance method. For example, you can also do this:
balance.transaction do balance.save! account.save! end
Transactions are not distributed across database connections
A transaction acts on a single database connection. If you have multiple class-specific databases, the transaction will not protect interaction among them. One workaround is to begin a transaction on each class whose models you alter:
Student.transaction do Course.transaction do course.enroll(student) student.units += course.units end end
This is a poor solution, but full distributed transactions are beyond the scope of Active Record.
Save and destroy are automatically wrapped in a transaction
Both Base#save and Base#destroy come wrapped in a transaction that ensures that whatever you do in validations or callbacks will happen under the protected cover of a transaction. So you can use validations to check for values that the transaction depends on or you can raise exceptions in the callbacks to rollback, including after_* callbacks.
Exception handling and rolling back
Also have in mind that exceptions thrown within a transaction block will be propagated (after triggering the ROLLBACK), so you should be ready to catch those in your application code.
Warning: one should not catch ActiveRecord::StatementInvalid exceptions inside a transaction block. StatementInvalid exceptions indicate that an error occurred at the database level, for example when a unique constraint is violated. On some database systems, such as PostgreSQL, database errors inside a transaction causes the entire transaction to become unusable until it’s restarted from the beginning. Here is an example which demonstrates the problem:
# Suppose that we have a Number model with a unique column called 'i'. Number.transaction do Number.create(:i => 0) begin # This will raise a unique constraint error... Number.create(:i => 0) rescue ActiveRecord::StatementInvalid # ...which we ignore. end # On PostgreSQL, the transaction is now unusable. The following # statement will cause a PostgreSQL error, even though the unique # constraint is no longer violated: Number.create(:i => 1) # => "PGError: ERROR: current transaction is aborted, commands # ignored until end of transaction block" end
One should restart the entire transaction if a StatementError occurred.
User.transaction do User.create(:username => 'Kotori') User.transaction do User.create(:username => 'Nemu') raise ActiveRecord::Rollback end end User.find(:all) # => empty
It is also possible to requires a sub-transaction by passing :requires_new => true. If anything goes wrong, the database rolls back to the beginning of the sub-transaction without rolling back the parent transaction. For example:
User.transaction do User.create(:username => 'Kotori') User.transaction(:requires_new => true) do User.create(:username => 'Nemu') raise ActiveRecord::Rollback end end User.find(:all) # => Returns only Kotori
Most databases don’t support true nested transactions. At the time of writing, the only database that we’re aware of that supports true nested transactions, is MS-SQL. Because of this, Active Record emulates nested transactions by using savepoints. See http://dev.mysql.com/doc/refman/5.0/en/savepoints.html for more information about savepoints.
If you’re on MySQL, then do not use DDL operations in nested transactions blocks that are emulated with savepoints. That is, do not execute statements like ‘CREATE TABLE’ inside such blocks. This is because MySQL automatically releases all savepoints upon executing a DDL operation. When #transaction is finished and tries to release the savepoint it created earlier, a database error will occur because the savepoint has already been automatically released. The following example demonstrates the problem:
Model.connection.transaction do # BEGIN Model.connection.transaction(:requires_new => true) do # CREATE SAVEPOINT active_record_1 Model.connection.create_table(...) # active_record_1 now automatically released end # RELEASE savepoint active_record_1 # ^^^^ BOOM! database error! end
@wiseleyb and @glosakti, neither of your suggestions are necessary, and both are bad practices.
test "transactions" do assert_raises ZeroDivisionError do User.transaction do 1/0 end end end
passes just fine on its own, with the transaction rolled back as you’d expect. No need to hack something ugly together.
Consider the following:
foo = Foo.new bar = Bar.new ActiveRecord::Base.transaction do foo.save! # succeeds bar.save! # failure, validation problem end foo.persisted? # true (!)
foo was not permanently stored in the database, but it was transiently saved, and this is reflected in the ActiveRecord model still in memory. But if you try
foo.reload # raises ActiveRecord::RecordNotFound
Don’t let stale data confuse you after using transactions!
Edited to add: This particular example does not succeed in reproducing the issue I encountered, which involved a slightly more complicated set of nested transactions. I haven’t managed to produce a simple test case where stale data remains in the model, but I have definitely experienced it in my app.
I think this is more appropriate.
def start_transaction Company.transaction do # don't forget the bang to make sure it raise # exception or the transaction won't rollback user.save! company.save! x=1/0 return true end # re-raise exception here raise "Exception!" end
Then you could call the method in another place, and it would raise rollback and other exception.
... # would return the "Exception!" if rollback occurs # it would also still trigger another exception other # than rollback. start_transaction ...