How to avoid race conditions when using the find_o

2019-04-30 03:25发布

From the documentation for find_or_create:

Note: Because find_or_create() reads from the database and then possibly inserts based on the result, this method is subject to a race condition. Another process could create a record in the table after the find has completed and before the create has started. To avoid this problem, use find_or_create() inside a transaction.

Is it enough to just use find_or_create() inside a transaction in PostgreSQL?

2条回答
对你真心纯属浪费
2楼-- · 2019-04-30 03:59

This implementation of find_or_create should prevent the race condition, described in the OP:

eval {
    $row = $self->model->create( { ... } );
}
if($@ && $@ =~ /duplicate/i) {
   $row = $self->model->find( { ... } );
} 

It also reduces find_or_create() to a single query in the best case.

查看更多
We Are One
3楼-- · 2019-04-30 04:19

No, the documentation is incorrect. Using a transaction alone does not avoid this problem. It only guarantees that the whole transaction is rolled back if an exception should occur - so that no inconsistent state will be persisted to the database.

To avoid this problem you must lock the table - inside a transaction, because all locks are released at the end of a transaction. Something like:

BEGIN;
LOCK TABLE mytbl IN SHARE MODE;

-- do your find_or_create here

COMMIT;

But that's not a magic cure for everything. It can become a performance problem, and there may be deadlocks (concurrent transactions mutually trying to lock resources that the other one has locked already). PostgreSQL will detect such a condition and cancel all but one of the competing transactions. You must be prepared to retry the operation on failure.

The PostgreSQL manual about locks.

If you don't have a lot of concurrency you might also just ignore the problem. The time slot is very tiny so it only very rarely actually happens. If you catch the duplicate key violation error, which will do no harm, then you have covered this, too.

查看更多
登录 后发表回答