Log-based wait strategy caused intermittent Testcontainers test failures for Keycloak

A developer debugging a flaky Testcontainers test for a custom Keycloak authenticator discovered that waiting on a startup log message did not guarantee the server was ready to accept HTTP connections. Keycloak logs its 'development mode' message milliseconds before its HTTP listener finishes binding, creating a narrow but real race condition. The test failed roughly one in three times with a connection reset error, depending on how fast the CI runner was. Replacing the log-based wait with an HTTP health check polling the actual endpoint resolved the flakiness entirely. The broader lesson drawn is to wait on the concrete side effect your code depends on — not on a log statement that signals intent rather than readiness.