Python SDET Interview Questions for Senior Level (2026)

Short answer: it depends on what the work waits for.

• asyncio - many network calls (seeding data, polling APIs) in one process.
• Threads - blocking I/O libraries that have no async version.
• Processes - CPU-heavy work, or full isolation between tests. pytest-xdist uses separate worker processes.

The reason: in standard CPython the GIL lets only one thread run Python bytecode at a time, so threads help with waiting, not with CPU work. (Python 3.13 added an optional free-threaded build, but it is experimental.)

Short answer: asyncio.gather for concurrency, a Semaphore to cap in-flight requests, and return_exceptions=True so you can report every failure, not just the first.

import asyncio
import httpx

async def create(client, sem, n):
    async with sem:
        r = await client.post("/users", json={"email": f"seed+{n}@example.com"})
        r.raise_for_status()
        return r.json()["id"]

async def seed(count=200, limit=10):
    sem = asyncio.Semaphore(limit)
    async with httpx.AsyncClient(base_url=BASE_URL, timeout=10) as client:
        results = await asyncio.gather(*(create(client, sem, n) for n in range(count)),
                                       return_exceptions=True)
    failures = [r for r in results if isinstance(r, Exception)]
    if failures:
        raise RuntimeError(f"{len(failures)} of {count} seeds failed: {failures[0]!r}")
    return results

ids = asyncio.run(seed())

Pick the limit with the environment owners; a shared staging server is not a load-test target.

Short answer: make every created resource unique per worker and per test, and never depend on run order. pytest-xdist exposes the worker id through the worker_id fixture ("gw0", "gw1"... or "master" without xdist).

import uuid
import pytest

@pytest.fixture
def user(api, worker_id):
    email = f"qa-{worker_id}-{uuid.uuid4().hex[:8]}@example.com"
    created = api.post("/users", json={"email": email}, timeout=10).json()
    yield created
    api.delete(f"/users/{created['id']}", timeout=10)

For one-time session setup shared across workers, the xdist docs suggest a file lock so only one worker performs it.

Short answer: separate layers so tests read like business steps and infrastructure can change underneath.

framework/
  clients/      # API clients, one per service (typed request/response models)
  pages/        # UI page objects
  data/         # builders and factories
  config/       # environment settings loaded from env vars
tests/
  api/  ui/  contract/
conftest.py     # fixtures wire the layers together

Rules worth stating: tests never build URLs or raw HTTP calls, config comes from environment variables (not code), and fixtures own setup and cleanup.

Short answer: implement the pytest_addoption hook in the root conftest.py and read it through request.config.getoption.

def pytest_addoption(parser):
    parser.addoption("--env", default="staging", choices=["dev", "staging"],
                     help="target environment")

@pytest.fixture(scope="session")
def base_url(request):
    return {"dev": "https://dev.example.com",
            "staging": "https://staging.example.com"}[request.config.getoption("--env")]

pytest --env dev -m smoke

Short answer: type hints plus a checker such as mypy catch wrong arguments before a slow test run. A Protocol lets tests depend on behaviour, so a fake client can replace the real one.

from typing import Protocol

class PaymentsClient(Protocol):
    def charge(self, order_id: str, amount_paise: int) -> str: ...

def checkout(client: PaymentsClient, order_id: str) -> str:
    return client.charge(order_id, 49_900)

class FakePayments:
    def charge(self, order_id: str, amount_paise: int) -> str:
        return "pay_test_123"

assert checkout(FakePayments(), "o1") == "pay_test_123"

Short answer: make flakiness visible, give it an owner, and fix causes.

1. Measure: record pass/fail per test across runs (JUnit XML into a dashboard) so you know which tests flip.
2. Quarantine: move confirmed flaky tests to a non-blocking job with a ticket and a deadline - not a permanent skip.
3. Fix the category: most causes fall into timing, shared data, order dependence or environment instability.
4. Prevent: review rules - no time.sleep, no shared mutable data, cleanup in fixtures.

Only quote improvement numbers you actually measured on your own suite.

Short answer: whether the test can fail for the right reason and only that reason.

• A clear assertion with a helpful message; no assertions hidden in fixtures.
• No time.sleep; waits are on a condition.
• Data created by the test and cleaned up by it.
• Network calls have timeouts.
• The test name states the behaviour, e.g. test_expired_coupon_is_rejected.