* generate_test_data.sh was spawning a brand-new kafka-console-producer JVM process per message (once/sec), paying full JVM startup + classpath scan + producer init/metadata fetch every iteration, forever — this pegged the osprey-kafka-test-data-producer container at 180-200% CPU cycling every 1-2s under docker top. * Now starts a single long-lived kafka-console-producer process and streams generated messages to it over a FIFO, instead of forking a new producer per line. * Added 'unless-stopped' restart policy to test data generator container Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>