Question 780
A dependency resolver silently upgrades pandas and breaks a tested vectorized UDF in production. How should the team avoid a repeat of this issue in production?
Explanation
Correct Answer
Use a reviewed lockfile or equivalent reproducible constraints. This is the most directly relevant preventive control because unconstrained transitive versions can introduce incompatible behavior between environments. A concrete acceptance check is: Rebuilt environments resolve the same versions and pass a representative UDF test. To reduce recurrence: Use a reviewed lockfile or equivalent reproducible constraints.
Incorrect Answers
Unbounded inclusion rules can deploy unintended assets with production privileges. For the observed behavior, unconstrained transitive versions can introduce incompatible behavior between environments.
Incremental maintenance capability depends on operator, source and dataset semantics. The controlling issue in this question is that unconstrained transitive versions can introduce incompatible behavior between environments.
Some relational operators cannot be maintained safely with streaming-table semantics. Here the important technical distinction is that unconstrained transitive versions can introduce incompatible behavior between environments.