Integrating Flyway with Java EE and using its datasource
Flyway is my favourite tool when it comes to database migrations because its convention over configuration approach makes it fairly easy to use while bootstrapping and configuration is reduced to a minimum. For those of you who don’t know anything about database migration tools (evolutionary database design) in short the concept is to track SQL scripts with your source code which reflect the actual version of your database that matches the code version to run properly. This could be done without any framework but what tools like Flyway or liquibase bring to the table is the ability to check automatically at build- or start-time if the database is at the latest available version and if not run all available SQL scripts from the actual database version to the newest one available. Flyway can be used with all the major SQL dialects, many different relational databases and has plugins for a huge variety of frameworks to be integrated into. For more detailed information visit the projects website.
As I’m a heavy Java EE user I ran into the problem that Java EE has no special support by Flyway. My problems with this are that there is no build in way to get the actual JPA datasource (defined in the persistence.xml) and there is no solution provided that wires Flyway in the startup process to run the migration scripts at startup. Flyway provides integration plugins with this functionalities for other frameworks like Spring Boot but fortunately it is not that complicated to realise the same thing in Java EE.
In many scenarios it is no problem to trigger Flyway in the right moment as it is possible to integrate it in the maven build process or start it manually via the command line but both solutions doesn’t fit my needs. I want to have Flyway check my database at startup as the production server is not the server where the build process happens and I want to avoid a situation where I have to execute a shell command manually after deploying a new version. For this cases Flyway can be started from the source code as well. As it has to be run before the application starts and tries using the database you have to make sure that the Flyway scripts are run before that. I found two solutions to make this work. The first one uses Hibernates “Service Provider Interface” (SPI) -I use WildFly in this example which comes with Hibernate- to register a new integrator and grab the datasource via a little bit of reflection code while the second approach (which is my favourited solution) defines a new singleton bean which uses @resource injection to get the datasource.