Death by CachingSpark · easy · ~10 minA well-meaning refactor added .cache() after nearly every transformation in the nightly ETL 'for performance'. Since then the job is slower, executors GC constantly, and two runs died with executor OOMs.Caching made things worse. Explain why, and set the caching policy.Evidence

Metrics

Evidence

Config

Checkpoints

Why did adding caches make the job slower and OOM-prone?

What's the correct caching policy for this job?

How do you verify the fix and explain the memory model interview-style?