Force loading of JDBC drivers in runtime classloader

[jaikiran]

Fixes #7079

The issue noted there is due to the way the java.sql.DriverManager deals with access to java.sql.Driver instances. Callers of various APIs on the DriverManager are allowed access to Driver based on classloader checks.

In that linked issue, it so happens that the Postgres driver gets loaded (early) in the augmentation classloader whereas the Tracing driver (the one which needs that underlying Postgres driver) gets loaded (a bit late) in the runtime classloader. As a result, the call from the Tracing driver, at runtime, is not allowed access to the Postgres driver.

The reason why this works in Java 11 is because in Java 11 the java.sql.DriverManager loads the drivers lazily[1] unlike in Java 8 where the DriverManager loads them eagerly[2]. Given the reliance on TCCL, in DriverManager for loading the drivers, this leads to potentially non-deterministic situation where the driver may or may not be loaded in the correct classloader.

To remedy this, the commit in this PR, forces loading of (any available) Drivers in the runtime classloader just before the datasource is being produced. This ensures that these drivers are then accessible in the rest of the runtime.

The commit also includes a testcase which reproduces the failure and verifies the fix.

[1] https://hg.openjdk.java.net/jdk/jdk/file/3b89be93a7e7/src/java.sql/share/classes/java/sql/DriverManager.java#l570
[2] https://hg.openjdk.java.net/jdk8u/jdk8u/jdk/file/4db0e91b95c8/src/share/classes/java/sql/DriverManager.java#l100

[stuartwdouglas]

Fix looks good to me, my only real question is why the test is in the opentracing module?

[jaikiran]

Hello @stuartwdouglas, thank you for the review.

my only real question is why the test is in the opentracing module?

I started off trying to reproduce the original issue which uses the opentracing the JDBC driver URL (which requires the opentracing extension). That way it tests the same code path of the original issue. Plus, I couldn't find any tests in our Quarkus codebase which tests the jdbc:tracing: JDBC URL format which is only mentioned in the docs https://quarkus.io/guides/opentracing#jdbc

Having said that, I think I can come up with a test case which reproduces this outside of opentracing. Let me know if you prefer that and I can work on it.