1. Overview
There are plenty of situations when we want to retrieve data from a database. Sometimes we want to lock it for ourselves for further processing so no one else can interrupt our actions.
We can think of two concurrency control mechanisms that allow us to do that: setting the proper transaction isolation level or setting a lock on data that we need at the moment.
The transaction isolation is defined for database connections. We can configure it to retain the different degree of locking data.
However, the isolation level is set once the connection is created, and it affects every statement within that connection. Luckily, we can use pessimistic locking, which uses database mechanisms for reserving more granular exclusive access to the data.
We can use a pessimistic lock to ensure that no other transactions can modify or delete reserved data.
There are two types of locks we can retain: an exclusive lock and a shared lock. We could read but not write in data when someone else holds a shared lock. In order to modify or delete the reserved data, we need to have an exclusive lock.
We can acquire exclusive locks using ‘SELECT … FOR UPDATE’ statements.
2. Lock Modes
JPA specification defines three pessimistic lock modes that we’re going to discuss:
- PESSIMISTIC_READ allows us to obtain a shared lock and prevent the data from being updated or deleted.
- PESSIMISTIC_WRITE allows us to obtain an exclusive lock and prevent the data from being read, updated or deleted.
- PESSIMISTIC_FORCE_INCREMENT works like PESSIMISTIC_WRITE, and it additionally increments a version attribute of a versioned entity.
All of them are static members of the LockModeType class and allow transactions to obtain a database lock. They all are retained until the transaction commits or rolls back.
It’s worth noting that we can obtain only one lock at a time. If it’s impossible, a PersistenceException is thrown.
2.1. PESSIMISTIC_READ
Whenever we want to just read data and don’t encounter dirty reads, we could use PESSIMISTIC_READ (shared lock). We won’t be able to make any updates or deletes, though.
It sometimes happens that the database we use doesn’t support the PESSIMISTIC_READ lock, so we could obtain the PESSIMISTIC_WRITE lock instead.
2.2. PESSIMISTIC_WRITE
Any transaction that needs to acquire a lock on data and make changes to it should obtain the PESSIMISTIC_WRITE lock. According to the JPA specification, holding PESSIMISTIC_WRITE lock will prevent other transactions from reading, updating or deleting the data.
Please note that some database systems implement multi-version concurrency control that allows readers to fetch data that has been already blocked.
2.3. PESSIMISTIC_FORCE_INCREMENT
This lock works similarly to PESSIMISTIC_WRITE, but it was introduced to cooperate with versioned entities — entities that have an attribute annotated with @Version.
Any updates of versioned entities could be preceded with obtaining the PESSIMISTIC_FORCE_INCREMENT lock. Acquiring that lock results in updating the version column.
It’s up to a persistence provider to determine whether it supports PESSIMISTIC_FORCE_INCREMENT for unversioned entities or not. If it doesn’t, it throws the PersistenceException.
2.4. Exceptions
It’s good to know which exception may occur while working with pessimistic locking. JPA specification provides different types of exceptions:
- PessimisticLockException indicates that obtaining a lock or converting a shared to exclusive lock fails and results in a transaction-level rollback.
- LockTimeoutException indicates that obtaining a lock or converting a shared lock to exclusive times out and results in a statement-level rollback.
- PersistenceException indicates that a persistence problem occurred. PersistenceException and its subtypes, except NoResultException, NonUniqueResultException, LockTimeoutException and QueryTimeoutException, mark the active transaction to be rolled back.
3. Using Pessimistic Locks
There are a few possible ways to configure a pessimistic lock on a single record or group of records. Let’s see how to do it in JPA.
3.1. Find
Find is probably the most straightforward way.
It’s enough to pass a LockModeType object as a parameter to the find method:
entityManager.find(Student.class, studentId, LockModeType.PESSIMISTIC_READ);
3.2. Query
Additionally, we can use a Query object as well and call the setLockMode setter with a lock mode as a parameter:
Query query = entityManager.createQuery("from Student where studentId = :studentId");
query.setParameter("studentId", studentId);
query.setLockMode(LockModeType.PESSIMISTIC_WRITE);
query.getResultList()
3.3. Explicit Locking
It’s also possible to manually lock the results retrieved by the find method:
Student resultStudent = entityManager.find(Student.class, studentId);
entityManager.lock(resultStudent, LockModeType.PESSIMISTIC_WRITE);
3.4. Refresh
If we want to overwrite the state of the entity by the refresh method, we can also set a lock:
Student resultStudent = entityManager.find(Student.class, studentId);
entityManager.refresh(resultStudent, LockModeType.PESSIMISTIC_FORCE_INCREMENT);
3.5. NamedQuery
@NamedQuery annotation allows us to set a lock mode as well:
@NamedQuery(name="lockStudent",
query="SELECT s FROM Student s WHERE s.id LIKE :studentId",
lockMode = PESSIMISTIC_READ)
4. Lock Scope
Lock scope parameter defines how to deal with locking relationships of the locked entity. It’s possible to obtain a lock just on a single entity defined in a query or additionally block its relationships.
To configure the scope, we can use PessimisticLockScope enum. It contains two values: NORMAL and EXTENDED.
We can set the scope by passing a parameter ‘jakarta.persistence’ with PessimisticLockScope value as an argument to the proper method of EntityManager, Query, TypedQuery or NamedQuery:
Map<String, Object> properties = new HashMap<>();
map.put("jakarta.persistence", PessimisticLockScope.EXTENDED);
entityManager.find(
Student.class, 1L, LockModeType.PESSIMISTIC_WRITE, properties);
4.1. PessimisticLockScope.NORMAL
The PessimisticLockScope.NORMAL is the default scope. With this locking scope, we lock the entity itself. When used with joined inheritance, it also locks the ancestors.
Let’s look at the sample code with two entities:
@Entity
@Inheritance(strategy = InheritanceType.JOINED)
public class Person {
@Id
private Long id;
private String name;
private String lastName;
// getters and setters
}
@Entity
public class Employee extends Person {
private BigDecimal salary;
// getters and setters
}
When we want to obtain a lock on the Employee, we can observe the SQL query that spans over those two entities:
SELECT t0.ID, t0.DTYPE, t0.LASTNAME, t0.NAME, t1.ID, t1.SALARY
FROM PERSON t0, EMPLOYEE t1
WHERE ((t0.ID = ?) AND ((t1.ID = t0.ID) AND (t0.DTYPE = ?))) FOR UPDATE
4.2. PessimisticLockScope.EXTENDED
The EXTENDED scope covers the same functionality as NORMAL. In addition, it’s able to block related entities in a join table.
Simply put, it works with entities annotated with @ElementCollection or @OneToOne, @OneToMany, etc. with @JoinTable.
Let’s look at the sample code with the @ElementCollection annotation:
@Entity
public class Customer {
@Id
private Long customerId;
private String name;
private String lastName;
@ElementCollection
@CollectionTable(name = "customer_address")
private List<Address> addressList;
// getters and setters
}
@Embeddable
public class Address {
private String country;
private String city;
// getters and setters
}
Let’s analyze some queries when searching for the Customer entity:
SELECT CUSTOMERID, LASTNAME, NAME
FROM CUSTOMER WHERE (CUSTOMERID = ?) FOR UPDATE
SELECT CITY, COUNTRY, Customer_CUSTOMERID
FROM customer_address
WHERE (Customer_CUSTOMERID = ?) FOR UPDATE
We can see that there are two FOR UPDATE queries that lock a row in the customer table as well as a row in the join table.
Another interesting fact to note is that not all persistence providers support lock scopes.
5. Setting Lock Timeout
Besides setting lock scopes, we can adjust another lock parameter — timeout. The timeout value is the number of milliseconds that we want to wait for obtaining a lock until the LockTimeoutException occurs.
We can change the value of timeout similarly to lock scopes, by using property ‘jakarta.persistence.lock.timeout’ with the proper number of milliseconds.
It’s also possible to specify “no wait” locking by changing timeout value to zero.
However, we should keep in mind that there are database drivers that don’t support setting a timeout value this way:
Map<String, Object> properties = new HashMap<>();
map.put("jakarta.persistence.lock.timeout", 1000L);
entityManager.find(
Student.class, 1L, LockModeType.PESSIMISTIC_READ, properties);
6. Conclusion
When setting the proper isolation level is not enough to cope with concurrent transactions, JPA gives us pessimistic locking. It enables us to isolate and orchestrate different transactions so they don’t access the same resource at the same time.
To achieve that, we can choose between discussed types of locks and consequently modify such parameters as their scopes or timeouts.
On the other hand, we should remember that understanding database locks is as important as understanding the mechanisms of underlying database systems.
It’s also important to remember that the behavior of pessimistic locks depends on the persistence provider we work with.
Lastly, the source code of this tutorial is available over on GitHub for Hibernate and for EclipseLink.