Python interview questions and answers

chain runs tasks sequentially, passing return values forward.

prefork (default) uses processes and is the safe choice for CPU and for C extensions that are not greenlet-safe.

Cap retries, catch expected exceptions, and route failures to a dead-letter queue or an error table.

A stampede is many requests missing the same expired key and all recomputing it at once.

Prefer versioned keys or a per-object cache generation stored on the model or in Redis so you bump a number instead of scanning keys.

UpdateCacheMiddleware/FetchFromCacheMiddleware store responses keyed by path plus the Vary-derived header values.

Use the cache framework for pickled Python objects with timeouts.

SimpleTestCase has no database and is for pure units and request factory tests that mock the ORM.

Set CELERY_TASK_ALWAYS_EAGER (or the modern task_always_eager) in test settings so delay() runs in-process.

@pytest.mark.django_db requests a test database and transaction rollback by default.

CPython's pymalloc arenas allocate many small objects itself and may not return that memory to the OS after frees.

Start tracemalloc early, snapshot before and after a request batch, and compare top statistics by traceback.

Evaluating a queryset stores the result list on the queryset object.

weakref.WeakValueDictionary drops entries when no strong references remain, so a memoizer does not pin models forever.

__getattribute__ runs on every attribute access and is easy to recurse infinitely if you are not careful

For a + b, Python tries a.__add__(b)

tee buffers items that one branch has consumed but the other has not, which can grow to the full stream.

They are lazy, so constructing the iterator looks cheap, but iterating can explode combinatorially and hang a worker.

Default ORM calls are blocking and must not run on the event loop.

gather can return exceptions if return_exceptions=True, otherwise the first exception fails the gather while sibling tasks may keep running unless you cancel them.