Relational database systems employ various Concurrency Control mechanisms to provide transactions with ACID property guarantees. While isolation levels are one way of choosing a given Concurrency Control mechanism, you can also use explicit locking whenever you want a finer-grained control to prevent data integrity issues.
As previously explained, there are two types of explicit locking mechanisms: pessimistic (physical) and optimistic (logical). In this post, I’m going to explain how explicit pessimistic locking interacts with non-query DML statements (e.g. insert, update, and delete).
For the upcoming test cases, we are going to use the following entities:
Post entity can have one or more
PostComment(s) which are associated with their parent entity through the
post_id Foreign Key.
Depending on the underlying database capabilities, a query can also acquire an exclusive (write) or a shared (read) lock on the database rows that are selected by the currently running statement. To see how the pessimistic locks interact with concurrent INSERT, UPDATE, and DELETE statements, we can use exclusive locks since they are better supported by most relational databases.
In the following examples, Alice is going to select all
PostComment(s) belonging to a given
Post entity while also acquiring an exclusive lock on the selected records. To simplify the lock acquisition, we can use the
LockModeType offered by Java Persistence API. Therefore, Alice query looks as follows:
List<PostComment> comments = session.createQuery( "select c " + "from PostComment c " + "where c.post.id = :id", PostComment.class) .setParameter("id", 1L) .setLockMode(LockModeType.PESSIMISTIC_WRITE) .getResultList();
This way, Hibernate is going to generate the database-specific locking clause on our behalf.
For instance, on Oracle and MySQL, the SQL query looks as follows:
SELECT pc.id as id1_1_, pc.post_id as post_id3_1_, pc.review as review2_1_ FROM post_comment pc WHERE pc.post_id = 1 FOR UPDATE
On PostgreSQL, an additional
OF alias is being employed as well:
SELECT pc.id as id1_1_, pc.post_id as post_id3_1_, pc.review as review2_1_ FROM post_comment pc WHERE pc.post_id = 1 FOR UPDATE OF pc
While on SQL Server, the syntax is a little bit different than the previous RDBMS:
SELECT pc.id as id1_1_, pc.post_id as post_id3_1_, pc.review as review2_1_ FROM post_comment pc WITH (UPDLOCK, ROWLOCK) WHERE pc.post_id = 1
However, you don’t have to worry about all these differences since Hibernate takes care of them for you.
Once the database records are locked, no UPDATE statement can modify them, even on a MVCC database engine.
Untile Alice releases the locks by ending her transaction, Bob’s UPDATE statement is blocked, and his transaction can no longer make any progress. After Alice has committed, the locks are released, and Bob’s transaction can continue.
Just like the UPDATE, a DELETE statement will also be blocked by the row-level locks acquired previously by Alice’s transaction:
Once the locks are released by Alice, Bob’s transaction can continue, and the DELETE is executed.
While UPDATE and DELETE statements are behaving consistently on most relational database systems (e.g. MySQL, PostgreSQL), the INSERT statement behaves differently.
Insert statement with MySQL
On MySQL, the explicit pessimistic locking clause, not only it acquires row-level locks on each record that is selected, but it can also acquire predicate locks as well because the InnoDB clustered index supports gaps and next-key locking.
This behavior can be observed only when using the default REPEATABLE READ isolation level. When switching to READ COMMITTED, MySQL behaves like PostgreSQL. One explanation would be that locks that did not match the scan are released after query execution. For more info, read this article on Percona blog.
So, Bob’s INSERT is blocked until Alice releases the predicate lock which was acquired on all (existing and future)
PostComment entries that are associated with a given parent
Insert statement with PostgreSQL
Unfortunately, explicit predicate locks are more an exception than a rule since most RDBMS don’t offer support for such a construct. For this reason, many relational databases behave like PostgreSQL:
Bob’s INSERT statement is executed right away even if Alice’s transaction tried to lock all
If you enjoyed this article, I bet you are going to love my book as well.
If you read my previous article about Phantom Read anomaly, and how it differs between 2PL and MVCC, you can draw same similarities between the Phantom Read anomaly prevention and the physical predicate lock support.
Therefore, on most RDBMS, explicit physical locks can only prevent record modifications for database records that existed at the time of locking while future records can be added. MySQL is an exception to this rule, preventing
INSERT statements against a range of locked entries.
So, the very same outcome you get from using
Serializable in regard to Phantom Reads, you also get by using explicit physical locking on a lesser isolation level (e.g. Read Committed). However, just like Two-pGase Locking, explicit locking has its price, so use it judiciously.