Whenever you're ready, walk me through how you'd approach this.
How do you handle and debug out-of-memory (OOM) errors in a distributed Spark cluster?
Explain a practical, systematic approach covering driver and executor memory, garbage collection, data skew, broadcast joins, shuffle spill, partition sizing, and Spark configuration. Include the diagnostic evidence you would collect, code or configuration changes you would test, and how you would validate the fix without masking data-quality or performance problems.