How do you use application debugging and introspection (inspect, IDE debuggers)?

Answer

IDE debuggers

Use breakpoints, step over/into/out, and inspect variables in the current frame. VS Code and PyCharm support Python debugging; attach to a process or run with “Debug” to hit breakpoints. For web apps, run the server in debug mode and trigger requests; the debugger stops in your route or service code. Use conditional breakpoints or logpoints when you need to stop only for certain inputs.

inspect module

inspect.getsource(obj) / inspect.getsourcefile(obj) — get source code or file of a function/class.
inspect.signature(func) — get parameter names and annotations.
inspect.getmro(cls) — get method resolution order for inheritance.
inspect.ismodule(), inspect.isfunction(), inspect.iscoroutine() — check object type.
inspect.stack() / inspect.currentframe() — get call stack; useful for logging or debugging “who called me.”
inspect.getmembers(obj) — list attributes and methods; useful for exploring unknown objects or generating docs.

pdb and breakpoint()

Insert breakpoint() (Python 3.7+) or import pdb; pdb.set_trace() to drop into an interactive debugger. Use n (next), s (step into), c (continue), p expr (print), l (list code), w (where/stack). Good for quick ad-hoc debugging when an IDE isn’t attached.

Logging and observability

Add structured logging (e.g. request ID, user, duration) at key points. Use log levels (DEBUG for dev, INFO/WARNING for prod). Optionally add tracing (OpenTelemetry) so you can trace a request across services and async tasks. Correlate logs with metrics and traces for incident investigation.

Profiling

Use cProfile or py-spy for CPU profiling to find bottlenecks. Use tracemalloc or memory_profiler for memory. Use asyncio debug mode or similar for async-specific issues (blocking calls, slow callbacks). Profile in conditions close to production (load, data size).

“How do you debug a production issue you can’t reproduce?” - start from telemetry: structured logs filtered by correlation ID, traces for where time went, and metrics for when it started. Reproduce locally only once you know the conditions.
“What do you reach for locally?” - breakpoint() for interactive inspection, py-spy for a live process without modifying it, tracemalloc for memory growth, and cProfile before optimising anything.
“How do you investigate a memory leak?” - tracemalloc snapshots compared over time, and check for the usual causes: unbounded caches, accumulating module-level state, and reference cycles holding objects with __del__.