Testing and debugging

1. "How do you find the real bug in a Java stack trace?"

Read to the last Caused by:. That is the root cause. Then open the first stack frame whose class is in your own package. The top exception is usually a wrapper. The JDK and library frames above your code are rarely the bug. ... N more stands for shared frames that the trace leaves out.

2. "Unit test vs integration test for a Kafka client?"

Unit-test the pure logic (parsing, tallying) with JUnit/AssertJ. These tests are fast and deterministic. Integration-test the actual produce/consume wiring against a real broker (native broker or Testcontainers). Don't integration-test what a unit test already covers.

3. "How do you grade a Kafka test without Docker?"

Use the native broker that starts from the Kafka tarball (a real JVM, no container). Or use an in-process TopologyTestDriver (Kafka Streams). Keep any Testcontainers/lab assertion as an extra layer that skips when Docker or the lab is unavailable. A skipped test cannot satisfy the "starter must fail" gate. So the native path must stand alone.

4. "Why assertThatThrownBy instead of a try/catch in the test?"

It captures the thrown Throwable from a lambda. It lets you fluently assert the type and the message. It fails with a clear message if the code throws nothing. A manual try/catch that does not fail on "no exception" is a classic too-weak test.

5. "Why not Thread.sleep while waiting for Kafka in a test?"

It is flaky if too short and slow if too long. Poll (consumer.poll in a loop) until you have the expected records or reach a deadline. This is deterministic and as fast as the data allows.

6. "What does a Java 21 NullPointerException tell you that older ones didn't?"

It has a helpful message that names exactly which reference was null (for example "because the return value of getX() is null"). You do not have to guess which dereference on the line failed.