Glossary

🎯 Interview angle

Reaching for !important usually signals a specificity problem rather than solving one. Methodologies like BEM exist to keep specificity flat.

🎯 Interview angle

Why z-index: 9999 sometimes does nothing - the element is trapped inside a context its ancestor created. An opacity of 0.99 is enough to create one.

🎯 Interview angle

border-box is what nearly every reset sets, because with the default content-box width: 100% plus padding overflows its container.

🎯 Interview angle

Surprises everyone once. Flex and grid containers do not collapse margins, which is one reason gap is easier to reason about.

🎯 Interview angle

Both respect the user's browser font-size setting, unlike px - that is the accessibility argument. rem avoids the compounding surprise.

🎯 Interview angle

Explains why height: 100vh overflows on mobile - vh historically ignored the address bar.

🎯 Interview angle

In InnoDB the primary key is the clustered index, and every secondary index stores the primary key rather than a row pointer - which is why a wide primary key inflates every other index.

🎯 Interview angle

EXPLAIN shows 'Using index'. The lookup back to the row is often the expensive part, so eliminating it is a bigger win than it sounds.

🎯 Interview angle

The leftmost-prefix rule is the whole question. An index on (a, b) helps WHERE a = ? and WHERE a = ? AND b = ?, but not WHERE b = ?.

🎯 Interview angle

WHERE YEAR(created) = 2026 scans; the date-range rewrite seeks. Same for a leading-wildcard LIKE. This is the most common accidental full scan.

🎯 Interview angle

It looks fine on ten rows and collapses on ten thousand. Django's fix is select_related (JOIN, single-valued) or prefetch_related (second query, multi-valued); catch regressions with assertNumQueries.

🎯 Interview angle

Name the anomaly each one prevents - dirty read, non-repeatable read, phantom - rather than reciting the ladder. Knowing your database's default matters too.

🎯 Interview angle

The systematic prevention is a consistent global lock ordering, which makes a wait-for cycle impossible. Beyond that: short transactions and retry logic.

🎯 Interview angle

The concrete bug is read-after-write: a user saves and the change appears to vanish. The fix is routing that read to the primary or pinning the session.

🎯 Interview angle

It scales reads only. Once the primary saturates on writes the next step is sharding, which is a far bigger change - saying that shows you know the ceiling.

🎯 Interview angle

Shard-key choice is the whole game. A key with skewed access creates a hot shard, and cross-shard joins and transactions become application problems.

🎯 Interview angle

'The key, the whole key, and nothing but the key.' Then say when you would denormalise - read-heavy paths where the join cost outweighs the duplication risk.

🎯 Interview angle

Do not claim MVT is fundamentally different from MVC. The honest framing is that Django names the layers differently and owns the routing.

🎯 Interview angle

Laziness is why chained filters cost nothing extra, and also why N+1 hides so well: the extra query fires at attribute access, far from where the queryset was defined.

🎯 Interview angle

Pair it with prefetch_related in your answer and say why they differ: a JOIN cannot express a multi-valued relation without row multiplication.

🎯 Interview angle

Two queries instead of N+1. Prefetch() lets you filter or order the prefetched set, which is the follow-up question.

🎯 Interview angle

Correct the old claim that the ORM is synchronous. Then add the nuance: these give an async-safe interface, but execute on a thread - not native async drivers.

🎯 Interview angle

Better than overriding save() because bulk operations and raw SQL bypass save() entirely. Same argument applies to db_default over default.

🎯 Interview angle

It covers inserts that never touch the ORM, and lets you use database functions such as Now(). A Python-side default cannot do either.

🎯 Interview angle

Order matters and is testable: AuthenticationMiddleware must follow SessionMiddleware because it reads the session. Describe it as an onion, not a list.

🎯 Interview angle

Say the drawback: implicit coupling that is hard to trace, and bulk_create / queryset.update() bypass save() so post_save never fires. An explicit service function is usually better.

🎯 Interview angle

A small detail that reliably distinguishes someone on a current Django from someone on 3.2.

🎯 Interview angle

The consequence is deployment: static can be collected and cached hard at build time, media needs writable, backed-up, usually off-server storage.

🎯 Interview angle

WSGI is one synchronous call per request with no concept of a persistent connection - which is why WebSockets need ASGI rather than being a bolt-on.

🎯 Interview angle

Say why: the old form burns the default slot on framework metadata and confuses type checkers. Annotated also lets you alias and reuse the dependency across routes.

🎯 Interview angle

The testing payoff is the strongest argument: dependency_overrides swaps a real database for a fake without touching application code.

🎯 Interview angle

Know the mapping cold: .dict() to .model_dump(), class Config to model_config = ConfigDict(...), orm_mode to from_attributes, @validator to @field_validator. Using v1 names is the clearest sign of a stale answer.

🎯 Interview angle

The filtering is a security feature - returning an ORM object with a password_hash is harmless if the response model does not declare it.

🎯 Interview angle

Setup before yield, teardown after. It keeps paired resources together and attaches shared state to the app rather than to globals.

🎯 Interview angle

They are complementary, not alternatives: Uvicorn serves ASGI, Gunicorn contributes process supervision and graceful restarts.

🎯 Interview angle

The single most common FastAPI performance bug. Use plain def (FastAPI runs it in a threadpool) or an async driver - async def around blocking code is the worst option.

🎯 Interview angle

A timeout never tells the client whether the write happened. Without idempotency the choice is duplicate charges or lost orders.

🎯 Interview angle

The distributed catch: an in-memory counter behind four instances enforces four times the intended limit. State belongs in Redis.

🎯 Interview angle

The first rule of ARIA is not to use it. A <button> is focusable and keyboard-activatable for free; role='button' on a div gives you the announcement and none of the behaviour.

🎯 Interview angle

Lead with accessibility, which is concrete and testable, rather than vague SEO benefit.

🎯 Interview angle

defer for application code that depends on the DOM or on ordering; async for independent third-party tags.

🎯 Interview angle

The orderable example is the test: promises resolve before setTimeout(0). A microtask that keeps scheduling microtasks starves the loop entirely.

🎯 Interview angle

Microtasks drain fully, macrotasks one at a time. That asymmetry explains most 'why did this log in that order' questions.

🎯 Interview angle

Behind private state, memoisation and the classic var-in-a-loop bug where every callback shares one binding. let creates a fresh binding per iteration.

🎯 Interview angle

let and const are hoisted, just not initialised. The TDZ converts a silent undefined into a loud error - that is the improvement over var.

🎯 Interview angle

Two wins: far fewer listeners, and it works for elements added after the listener was attached. Pair it with event.target.closest().

🎯 Interview angle

stopPropagation halts the rest of the journey, which silently breaks delegated handlers elsewhere - worth flagging as a cost.

🎯 Interview angle

Match to intent: debounce for 'user finished typing', throttle for 'keep updating while scrolling'. Choosing wrongly is the common error.

🎯 Interview angle

The browser enforces it, not the server - which is why curl is unaffected and the request often runs server-side regardless. Wildcard origin plus credentials is forbidden by spec.

🎯 Interview angle

The security trade is the answer: a token in localStorage is readable by any XSS; an HttpOnly cookie is not, but then needs CSRF protection.

🎯 Interview angle

It needs ESM because require() can be dynamic. Module side effects also block it - that is what sideEffects: false declares.

🎯 Interview angle

Usually the largest front-end performance lever, dominating almost any micro-optimisation. Code splitting and lazy loading are the tools.

🎯 Interview angle

Do not stop at 'Python can't do parallelism'. Say: on a default build the GIL applies, so use processes for CPU-bound work. Then add that the free-threaded build (PEP 703) removes it - experimental in 3.13, officially supported in 3.14 - and that removing it does not make your own data structures thread-safe.

🎯 Interview angle

Three things show you have actually followed it: it is a distinct build not a flag, single-threaded code is somewhat slower because refcounting must become thread-safe, and C extensions must opt in.

🎯 Interview angle

A classic screening question. Use None as the default and build inside the body. Dataclasses raise at class-creation time instead, which is why they force field(default_factory=list).

🎯 Interview angle

a.copy(), list(a) and {**d} are all shallow. Mutating a nested list shows through both copies. copy.deepcopy() fixes it and is far more expensive - know both facts.

🎯 Interview angle

The consequence is what gets asked: assigning to a name anywhere in a function makes it local for the whole function, so reading it earlier raises UnboundLocalError rather than falling through to the global.

🎯 Interview angle

Behind decorators, memoisation and the late-binding loop bug where every closure sees the final loop value. The fix is a default argument or functools.partial.

🎯 Interview angle

Always mention functools.wraps - without it the decorated function loses its name and docstring, breaking introspection and documentation tooling.

🎯 Interview angle

A memory optimisation for classes instantiated in the millions. The costs: no ad-hoc attributes, and multiple inheritance with slots gets awkward.

🎯 Interview angle

It guarantees a class precedes its parents and preserves base order. When no consistent order exists Python refuses to create the class - that is the 'cannot create a consistent MRO' error.

🎯 Interview angle

The machinery behind property, classmethod, staticmethod and ORM fields. Knowing that property is a descriptor is what turns it from magic into mechanism.

🎯 Interview angle

The memory argument is the point: constant memory regardless of length, which is how you stream a file larger than RAM. The trade is single-consumption and no indexing.

🎯 Interview angle

Distinguish iterable from iterator. A list is iterable and gives a fresh iterator each time; a generator is its own iterator and exhausts once.

🎯 Interview angle

It is try/finally with a name. __exit__ runs even on exception, and returning True from it suppresses the exception - a detail worth knowing but rarely worth using.

🎯 Interview angle

Mentioning it signals you have used a recent Python. The older TypeVar form still works everywhere and is what you will meet in existing codebases.

🎯 Interview angle

Static duck typing. The contrast with ABC is the answer: ABC requires explicit inheritance and enforces at instantiation; Protocol is checked by the type checker and requires nothing at runtime.

🎯 Interview angle

Concurrency, not parallelism - one thread, one core. The killer detail: one blocking call inside a coroutine stalls every other task on the loop.

🎯 Interview angle

The common bug is forgetting to await, which produces a 'coroutine was never awaited' warning and silently does nothing.

🎯 Interview angle

The virtual DOM is not faster than hand-written DOM code. Its value is that you describe target state declaratively and never manage transitions.

🎯 Interview angle

Index keys break on reorder or insertion: component state follows position, not item, so a checked checkbox jumps rows after a deletion.

🎯 Interview angle

The classic 'why is my counter stuck at 0'. The effect closes over the render that created it; the deps array is what forces recreation.

🎯 Interview angle

Explains why memo often does nothing: an inline arrow or object literal is a new reference every render. Also why mutating state in place never re-renders.

🎯 Interview angle

Mismatches come from non-determinism in render: Date.now(), Math.random(), localStorage. Move those into useEffect so the first client render matches the server.

🎯 Interview angle

Trade both ways: better first paint and crawlability, at the cost of server load and hydration. CSR ships an empty div and a bundle.

🎯 Interview angle

It exists because cookies are sent automatically. Token-based auth in a header is not vulnerable in the same way - which is the trade against HttpOnly cookies.

🎯 Interview angle

Context-aware output escaping is the fix, not input filtering. Django templates and Flask's Jinja2 setup autoescape by default - so the risk lives in |safe and dangerouslySetInnerHTML.

🎯 Interview angle

SHA-256 is wrong precisely because it is fast. Password hashes need a tunable work factor so brute force stays impractical.

🎯 Interview angle

State the cost: you cannot easily revoke one before it expires. Short-lived access tokens plus refresh tokens is the standard mitigation.

🎯 Interview angle

The current default answer for dependency management. Say what it consolidates and mention poetry as the common incumbent - that shows landscape awareness rather than tool fashion.

🎯 Interview angle

Know the boundary: Ruff does no type inference, so mypy or pyright stays in the stack. Fast enough to run in a pre-commit hook without friction.

🎯 Interview angle

The problem it solves is version conflict between projects, not security or performance. uv manages them without an explicit activate step.

🎯 Interview angle

Python type hints are not enforced at runtime - that is the whole point of a separate checker. Adopting gradually with per-module strictness is the realistic answer.

🎯 Interview angle

Explicit injection over hidden setUp state, plus scopes (function, module, session) so expensive setup is not repeated per test.