VocalHyre
Sign up

Python · The language, not just syntax

Which Python profiling tools do you use and how do you read them?

HardprofilingcprofilebenchmarkingUpdated Oct 2026
Quiz this track

Short answer

Use cProfile for deterministic function-level profiles, timeit for microbenchmarks, line_profiler for hot lines, and py-spy for sampling production processes without restarting them.

Deterministic profilers instrument calls and record every function invocation, which gives precise call counts and cumulative time but adds overhead. Sampling profilers periodically capture stacks, which is lighter and better for live processes.

python
import cProfile
import pstats

profiler = cProfile.Profile()
profiler.enable()
run_job()
profiler.disable()

stats = pstats.Stats(profiler)
stats.sort_stats("cumulative").print_stats(15)
Cumulative time shows which call paths are expensive overall. Self time shows where the work actually happens.

Reading a profile

  • tottime is time in the function itself; cumtime includes callees
  • High call counts with low time may indicate overhead rather than a target
  • A function can be hot because of its own work or because of what it calls
  • Use timeit for small isolated comparisons and repeat runs to reduce noise

Benchmark the operations that match production: same data size, same concurrency and same storage. A microbenchmark of a dict lookup rarely predicts end-to-end latency. Report percentiles, not just averages, because tail latency drives user experience.

Profiler overhead can distort allocation and I/O behaviour. Confirm a suspected hotspot with a second method, such as py-spy or targeted timing, before restructuring code.

Follow-ups you should expect

Interviewers use these to check whether the first answer was understanding or memorisation.

  1. 1What is the difference between tottime and cumtime?
  2. 2When would you use a sampling profiler instead of cProfile?
  3. 3How do you profile a running production process safely?
  4. 4Why are averages misleading in performance reports?

Say it in the interview

I use cProfile for deterministic function-level data, timeit for isolated comparisons, and py-spy when I need a sampling profile of a live process. I look at cumulative time to find expensive paths and self time to find where work happens, and I confirm hotspots with a second method. I benchmark realistic data sizes and report percentiles, because averages hide the tail latency that users actually feel.

Use this as a model, not a script. Saying it in your own words is what makes it survive follow-up pressure.

Related questions

HardHow do you optimise Python code without wasting effort?HardHow does Python manage memory and how do you reduce usage?MediumHow does Python's garbage collector handle cycles?

Reading is step one

Now say it out loud.

Understanding an answer and delivering it under pressure are different skills. Rehearse this topic in a live Python mock interview and get scored feedback on depth, clarity and confidence.

Start a free Python interviewBrowse all questions