Liferay - Multi-Stage Development and Data Management

Our development environments and data migration strategy is something that has evolved during the project, and will continue to evolve during development.  Process and design improvements are encouraged and investigated in an attempt to optimize the end product.

When multiple developers are involved in a large project, multi-staging is extremely important.  It allows developers to work simultaneously and independently, without affecting each other.  It also allows stability and loss testing along with a bunch of advantages. The number of stages required should depend on several things (this is obviously not a complete list):
  • Size of the project
  • Complexity
  • Number of developers
Two stages is the minimum required for any project, one for development and one for production.  However, it's highly recommended to have a Test environment between Development and Production.  The Test environment should be identical, or at least as close to the Production development as possible to minimize risk when deploying updates to Production.

In our situation we have three stages, which are:
  • Development
  • Test
  • Production
Furthermore, we are currently debating adding a fourth stage, QA, that would be always kept identical to the Production environment.

To meet our requirements, the Production environment is designed to be highly available with session replication.  Each stage incrementally becomes more similar to the Production environment.  The incremental changes spread out the issues that are due to environment variations, easing debugging.  Each environment is designed specifically for a purpose. 

Initial development is performed on a local desktop or laptop, which is not listed above.  The local environment is the most flexible, easily restarted and best for independent development and debugging.  Each developer has a complete environment running locally on their desktop/laptop.  This environment, which is used for rapid development, initial integration and testing by the developer, debugging, etc...  It gives developers the maximum freedom to work without affecting each other, which is very important near the start of the project because restarting services is quite common.  Locally we're developing using Liferay Developer Studio on Microsoft Windows.

We're using Development for initial integration between developers and testing.  It is running on RHEL.  The Test environment adds session replication, is located in a DMZ with public access and uses an Oracle database.  The differences in the development and great environment increase reliability and performance.  They also make the test environment extremely similar to the Production environment.

Data migration between environments quickly became important to synchronize
configurations and reduce overall effort.  We use built in features of Liferay to migrate documents, content and configurations.

To migrate web content as well as documents, we export and import LAR (Liferay ARchive) files. They're easy to use, however sometimes we notice issues with improper migration and have to repeat the export and/or import.  Issues we've encountered include missing content and permissions.

You can use staging to migrate pages and other configuration, however we we are using the database migration (Control Panel -> Server Administration -> Data Migration)  in Liferay. Data migration is a bit more work to use, but right now we aren't synchronizing individual pages -- which is the main advantage of staging -- we generally want to synchronize everything.

The headaches that go along with database migration are having to initialize the target database (shutting down the instance of Liferay that's using the database, and deleting all tables within the database) and then restarting both Liferay instances.  It is much more powerful though, allowing us to migrate from any environment to another. We find that it is extremely useful to migrate from Development or Test to our local environment during integration and testing.  It will also be important to migrate from production to test, development or locally when debugging and trying to recreate problems that occur in production.

To aid with data migration we switched from using the default hypersonic database to using MySQL in all the environments before test.  The Test and Production environments are using an Oracle database.  Using mysql was done for several reasons. 
  1. Allows database backups and quick restores because the database does sometimes become corrupt. We have nightly backups of our local and development databases.
  2. Allows database migration to and from any environment
  3. Allows direct access to the database for debugging and clearing lock records (Lock_ table)
Originally we were hosting Liferay on Glassfish, but we ran into several issues including session replication and data migration. Since then we have switched to using tomcat as our application server, and have not had any application server problems since.

So for now, we have a pretty complete development strategy with respect to staging, backups and data migration. We did not arrive at this state right away. Several of the decisions were made during development in an attempt to streamline processes, reduce effort, improve efficiency and end up with a more easily maintainable end product; something that our team is constantly working to achieve.

No comments: