Pair Programming: Debug and Extend a Live Codebase Collaboratively Under Time Pressure
Interview Experience
Problem
The interviewer shares a partially implemented task scheduler (roughly 150 lines of Python). It has 3 known failing tests and 2 undocumented bugs. You will pair-program live: read the code, diagnose failures, fix them, and add a new cancel_task feature — all within 45 minutes.
The scheduler interface:
python
class TaskScheduler:
def schedule(self, task_id: str, delay_sec: int, fn: Callable) -> None: ...
def cancel(self, task_id: str) -> bool: ...
def get_pending(self) -> list[str]: ...
Failing tests hint at:
- A race condition when two tasks are scheduled for the same time
- cancel returning True for already-executed tasks
- get_pending including tasks that errored during execution
Follow-ups
- How do you approach reading unfamiliar code quickly — what do you look at first?
- The race condition is in a heap push/pop. How do you make the heap thread-safe?
- How would you add retry logic with exponential backoff for failed tasks?
- If you needed to persist the task queue across restarts, what would you change?
Full Details
Problem
The interviewer shares a partially implemented task scheduler (roughly 150 lines of Python). It has 3 known failing tests and 2 undocumented bugs. You will pair-program live: read the code, diagnose failures, fix them, and add a new cancel_task feature — all within 45 minutes.
The scheduler interface:
python
class TaskScheduler:
def schedule(self, task_id: str, delay_sec: int, fn: Callable) -> None: ...
def cancel(self, task_id: str) -> bool: ...
def get_pending(self) -> list[str]: ...
Failing tests hint at:
- A race condition when two tasks are scheduled for the same time
- cancel returning True for already-executed tasks
- get_pending including tasks that errored during execution
Follow-ups
- How do you approach reading unfamiliar code quickly — what do you look at first?
- The race condition is in a heap push/pop. How do you make the heap thread-safe?
- How would you add retry logic with exponential backoff for failed tasks?
- If you needed to persist the task queue across restarts, what would you change?