A Spark stage has one long-running task while most tasks finish quickly. What should the engineer investigate?

1. Data skew and uneven partition sizes
2. Only notebook cell colors
3. Assume total row count must be zero
4. Reduce logging before reviewing task metrics