Blogs

Here I am again with another in the series about Liferay's architecture. If you haven't read them yet, the four previous entries covered: Overview, Services Layer, Web Services and Service Builder.

This time I'm going to cover a very important concept: caching. In today's web, it's impossible for a web application to provide even good enough performance in the web unless it has a well designed caching system. So what I'm going to cover here is not only useful to understand Liferay's architecture better but might also be beneficial for anyone writting Java web applications, specially if they are large.

Liferay is known to provide very good performance (check the Performance Whitepaper in the whitepapers section of this website for details) and its caching system is a key factor in achieving that performance. This caching system spans through all three layers. The following diagram shows the main caching components in each layer:

Let's cover each of them one by one, starting with the lower layer.

At the persistance layer Liferay relies on Hibernate to do most of its database access. Hibernate has two cache layers called Level 1 (L1) and Level 2 (L2). Level 1 is used to cache objects retrieved from the database within the current datababase session. In the case of Liferay a session is tied to an invocation to a service layer. So when the frontend layer (or a web service) invokes a service a db session is opened and reused until the service method returns. All operations performed until that point will share the L1 cache so the same object won't be retrieved twice from the database.

Hibernate's cache Level 2 is able to span across database sessions and stores database objects (Entity Cache) and results of queries (Query Cache). For example if any logic retrieves from the database all users that belong to a certain organization, the result will be stored in the cache. Note that what is really stored is the list of "references" to the actual objects which are stored in the Entity Cache. This is done to ensure that there aren't several copies of the same object in the cache.

Besides using Hibernate, Liferay's code also performs some complex database queries directly (although reusing the database conneciton). For these queries Liferay has its own Entity and Query cache. In fact thanks to the work of Shuyang these two caches are extremely efficient and as a result, for Liferay 6.2, we have decided to disable Hibernate's level 2 cache by default leaving Liferay's Query cache as responsible for that task to improve performance.

One final important aspect is that all of these caches use an underlying cache provider to manage the objects in memory. By default Liferay uses ehcache to do that, but it is also possible to change the caching provider through portal.properties.  The configuration is done within hibernate-clustered.xml which defines  and configures several cache areas.

Finetunning these configuration files is a very important task for any high-profile site. But don't try to do it based on guesses. You should always do it while running a performance test suite that can help you find bottlenecks and verify changes in the configuration.

Since I want to try and keep these blog entries shorter and more frequent, I'm dividing it in two parts. In the next one I'll cover the caching mechanisms of the services and the frontend layer.