Optimistic Locking by concrete (Java) example

Question

I have spent my morning reading all the top articles that Google churns up on optimistic locking, and for the life of me, I still don\'t really get it.

I underst

Answer 1

Normally when you look into optimistic locking you also use a library like Hibernate or an other JPA-Implementation with @Version support.

Example could read like this:

public class Person {
    private String firstName;
    private String lastName;
    private int age;
    private Color favoriteColor;
    @Version
    private Long version;
}

while obviously there is no point of adding a @Version annotation if you are not using a framework which supports this.

The DDL could then be

CREATE TABLE people (
    person_id PRIMARY KEY AUTO_INCREMENT,
    first_name VARCHAR(100) NOT NULL,
    last_name VARCHAR(100) NOT NULL,        # } I realize these column defs are not valid but this is just pseudo-code
    age INT NOT NULL,
    color_id FOREIGN KEY (colors) NOT NULL,  # Say we also have a colors table and people has a 1:1 relationship with it
    version BIGINT NOT NULL
);

What happens with the version?

1. Every time before you store the entity, you check if the version stored in the database is still the version you know.
2. If it is, store your data with version incremented by one

To get both steps done without risking an other process changing data between both steps it is normally handled through a statement like

UPDATE Person SET lastName = 'married', version=2 WHERE person_id = 42 AND version = 1;

After executing the statement you check if you updated a row or not. If you did, nobody else changed the data since you've read it, otherwise somebody else changed the data. If somebody else changed the data you will normally receive an OptimisticLockException by the library you are using.

This exception should cause all changes to be revoked and the process of changing the value to be restarted as the condition upon which the entity was to be updated may no longer be applicable.

So no collision:

1. Process A reads Person
2. Process A writes Person thereby incrementing version
3. Process B reads Person
4. Process B writes Person thereby incrementing version

Collision:

1. Process A reads Person
2. Process B reads Person
3. Process A writes Person thereby incrementing version
4. Process B receives an exception when trying to save as the version changed since Person was read

If Colour is another object you should put a version there by the same scheme.

What isn't Optimistic Locking?

• Optimistic Locking is no magic to merge conflicting changes. Optimistic Locking will just prevent processes from accidentally overwriting changes by another process.
• Optimistic Locking actually is no real DB-Lock. It just works by comparing the value of the version column. You don't prevent other processes from accessing any data, so expect that you get OptimisticLockExceptions

What column-type to use as version?

If many different applications access your data you may be best off using a column automatically updated by the database. e.g. for MySQL

version TIMESTAMP DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP;

this way the applications implementing optimistic locking will notice changes by dumb applications.

If you update entities more often than the resolution of TIMESTAMP or the Java-interpretation of it, ths approach can fail to detect certain changes. Also if you let Java generate the new TIMESTAMP you need to ensure that all machines running your applications are in perfect time-sync.

If all of your applications can be altered an integer, long, ... version is normally a good solution as it will never suffer from differently set clocks ;-)

There are other scenarios. You could e.g. use a hash or even randomly generate a String every time a row is to be changed. Important is, that you don't repeat values while any process is holding data for local processing or inside a cache as that process will not be able to detect change by looking at the version-column.

As a last resort you may use the value of all fields as version. While this will be the most expensive approach in most cases it is a way to get similar results without changing the table structure. If you use Hibernate there is the @OptimisticLocking-annotation to enforce this behavior. Use @OptimisticLocking(type = OptimisticLockType.ALL) on the entity-class to fail if any row changed since you have read the entity or @OptimisticLocking(type = OptimisticLockType.DIRTY) to just fail when another process changed the fields you changed, too.

Answer 2

@TheConstructor: Great explanation of what it is and is not. When you said "Optimistic Locking is no magic to merge conflicting changes" I wanted to comment. I used to manage a DataFlex application which allowed users to edit records on a form. When they pushed the "Save" button, the app would do what was called a "multi-user re-read" of the data - pulling the current values - and compare to what the user had modified. If the fields the user modified had not been changed in the meantime, then just those fields would be written back to the record (which was locked only during the re-read + write operation) and therefore, 2 users could transparently modify different fields on the same record with no issues. It did not require version stamps, just a knowledge of which fields were modified.

Granted, this is not a perfect solution, but it did the job in that case. It was optimistic and it did allow unrelated changes and it gave the user an error on conflicting changes. This was the best that could be done, pre-SQL, but is still a good design principle today, perhaps for more object- or web-related scenarios.