Python Interview Questions for Data Engineers

01

The 5 kinds of Python question

Cut by what a question asks you to DO, not by the topic it names. The same topic scores differently when you have to explain it, build it, or work out why it broke — and those three fail separately.

• 25ConceptualThe questions with no code attached. Why a generator is not a list, what a context manager guarantees that a try/finally does not, and when the GIL is the thing that is actually slowing you down.Comprehensions & generators · Decorators & context managers · GIL & memory · Data structures & complexity→
• 26Approach & Data StructuresA dict of lists against a defaultdict against a Counter; a generator against a list; a dataclass against a tuple. Given the shape and the size of the data, which do you reach for and what does the wrong choice cost?Data structures & complexity · Comprehensions & generators · OOP & dataclasses · GIL & memory→
• 23Output PredictionWhat does this print? Drawn from the language corners that trip people who use Python daily — a mutable default argument, a closure over a loop variable, a generator consumed twice, a shallow copy shared between rows.Comprehensions & generators · Data structures & complexity · OOP & dataclasses · GIL & memory→
• 23Debug & FixCode that fails, or worse, code that succeeds with the wrong answer. Find it and fix it: the swallowed exception, the retry that retries a non-idempotent write, the encoding assumption that holds until one file does not.Error handling & retries · File I/O & formats (CSV/JSON/Parquet) · Testing & mocking · Concurrency (threading vs multiprocessing vs asyncio)→
• 22Code Review & RefactorWorking code that would not survive a production incident. Make it reviewable: name the contract with types, make the failure observable with logging, make the retries safe, and make the whole thing testable without a live warehouse.Typing · Logging · Testing & mocking · Packaging & envs→

02

The ones that keep coming up

Answered here, in short. The full reasoning, the wrong answer and the follow-up are one click away.

1. 01What is the difference between mutable and immutable types in Python, and name a bug that follows directly from getting it wrong.Immutable objects cannot be changed after creation — int, float, str, tuple, frozenset. Mutable ones can — list, dict, set, and most class instances. The distinction decides what happens when an object is shared, defaulted or used as a key.→
2. 02What does the @dataclass decorator actually generate for you, and what does it deliberately not do?It reads the annotated class attributes and writes __init__, __repr__ and __eq__ from them, in declaration order. It does not validate types, and it refuses a mutable default outright.→
3. 03What does __slots__ do, and when is it worth adding to a class in a data pipeline?It declares a fixed set of attributes and drops the per-instance __dict__, which saves real memory and makes attribute access slightly faster. It is worth it when you hold millions of instances, and pointless below that.→
4. 04What does it mean for an object to be hashable, and what is the contract between __eq__ and __hash__?Hashable means it has a __hash__ that never changes and an __eq__ that agrees with it: equal objects must hash equal. Containers find objects by hash first, so an object whose hash could change would become unfindable.→
5. 05What is the difference between an iterable and an iterator, and what would you have to write to implement one by hand?An iterable has __iter__ and can produce a fresh iterator on demand; an iterator has __iter__ and __next__ and holds a position, so it is consumed once. A list is iterable; a generator is an iterator.→
6. 06Merge 500 CSV files from a directory into one pandas DataFrame. What is the mistake almost everyone makes on the first attempt?Read each file into a list and call pd.concat once at the end. Concatenating inside the loop copies the whole accumulated frame on every iteration, which is quadratic.→
7. 07You are summing millions of monetary amounts and converting between currencies. Why not float, and what do you use instead?Binary floats cannot represent most decimal fractions, so errors accumulate over millions of additions. Use integer minor units, or Decimal with an explicit context — and never mix the two.→
8. 08Compare two ledgers in memory and return what is missing on each side and what disagrees. How do you structure it, and what does it cost?Index both sides into dicts by the business key, then take the three set operations on the key sets: only-in-A, only-in-B, and in-both-where-values-differ. Linear time, and the memory is one dict.→
9. 09You have to process a 200 GB file on a machine with 16 GB of RAM. How do you approach it?Stream it: read line by line or in chunks, keep only bounded state, and write output incrementally. If the computation genuinely needs all the data at once, partition it so each part fits, or move it to an engine built for that.→
10. 10A team is choosing between CSV, JSON, NDJSON, Parquet and Avro for a new dataset. How do you decide?Decide by how it is read and who has to read it: columnar (Parquet) for analytics over a few columns of many rows, NDJSON for streaming appends and nested payloads, Avro when schema evolution is a first-class requirement, CSV only when a human or a legacy tool demands it.→
11. 11When do you convert a record from a plain dict to a dataclass, and when is a dict the right answer?Dicts at the edge, where the shape is unknown and unvalidated; typed records inside, once the keys are a contract other code relies on. Convert once, at the validation boundary, and stop using dicts after it.→
12. 12A config dict is copied with dict(template), a nested list inside it is appended to, and the template changes too. Why?dict() is a shallow copy. It builds a new outer dict whose values are the same objects as the original's, so every nested list and dict is still shared. Use copy.deepcopy, or rebuild the nested parts.→
13. 13Two @dataclass instances built from the same values are compared with == and with is. What does each return, and what changes with frozen=True?== is True because @dataclass generates a field-by-field __eq__; is is False because they are two objects. frozen=True adds a __hash__, which is what a plain dataclass lacks and why it cannot be a set member.→
14. 14Rows are sorted by amount, and two runs over the same data return the tied rows in a different order. Python's sort is stable — how is that possible?Stable means equal elements keep their input order. It says nothing about what the input order was — and if the rows came from a set, a dict rebuild or a parallel read, that order is not a property of your data.→
15. 15Every log line and stack trace in a service suddenly says "wrapper" instead of the function that failed. What happened?A decorator returned an inner function without copying the wrapped function's metadata. Add @functools.wraps(fn) to the wrapper and the name, docstring and module come back.→
16. 16A helper collects rejected rows and returns them. The first batch reports 12 rejections, the second reports 27, and the numbers only ever go up. Why?The function has a mutable default argument. The default list is created once when the function is defined, so every call that omits the argument appends to the same list for the life of the process.→
17. 17A finance report built in Python disagrees with the warehouse by a few pence per thousand rows. Nothing errors. Where do you look?Two causes, usually both: money held as float, and round() using banker's rounding where the specification says half-up. Fix by parsing to Decimal from the source string and quantizing with an explicitly named mode.→
18. 18A code review turns up a class with an __init__ that stores three arguments and exactly one other method. What would you say?If it has no state that outlives a call and only one method, it is a function with extra ceremony. Make it a function taking those arguments — unless the constructor is doing expensive setup that several calls share, which is the case where the class earns its place.→
19. 19A function branches on isinstance across four types to decide how to write records. What would you change, and what does it cost?Type the parameter as a Protocol describing the methods you actually call, and let anything with them satisfy it. The ladder has to be edited for every new implementation, and it rejects a perfectly valid object for not being in the family tree.→

03

What Python interviews actually cover

• 01Data structures & complexity
• 02Comprehensions & generators
• 03OOP & dataclasses
• 04Decorators & context managers
• 05Error handling & retries
• 06File I/O & formats (CSV/JSON/Parquet)
• 07pandas / polars
• 08APIs & pagination
• 09Concurrency (threading vs multiprocessing vs asyncio)
• 10GIL & memory
• 11Testing & mocking
• 12Typing
• 13Logging
• 14Packaging & envs

04

Learn it, then practise it