CAP theorem
~11 minTwo data centres hold the same data. The link between them breaks. Each must now answer from its own copy, or refuse. CAP is the proof that there is no third way.
Prepare for full stack JavaScript interviews, from React and Next.js in the browser to the Node.js services behind them. Covers TypeScript, Express and NestJS, data access, testing and system design.
14918 of 14918 questions in your plan: 0 known, 0 in progress, 0 due, 0 skipped, 14918 new. 0 cut, outside your plan.
A JavaScript full stack loop tests one language in two runtimes. The screen and the coding round are in JavaScript or TypeScript, and many teams swap the algorithm problem for a slice of the stack itself, a React view over an Express or NestJS endpoint, written live or set as a take-home. Technical conversations on the runtime, React and the server follow. From middle level up a design round is added, its brief as likely to be a feed or a typeahead as a backend service, and a behavioral interview closes. Short snippets carry most of the technical conversation, and several of them only make sense with both runtimes in view. A zero-delay `setTimeout` and an already resolved promise: which callback runs first, and where does `process.nextTick` fall relative to both? Why does one synchronous loop over a large array inside a route handler stall every other request on that process? Why does an effect with an empty dependency array keep reading the count from its first render? A `Date` leaves the server as a `Date` and arrives in the browser as a string, while the type both sides import still promises a `Date`: what checks that payload, if the compiler does not? A junior is expected to make the component and the endpoint work and to predict what those snippets do. Seniors are judged on placement: which process a line runs in, and what moving it would cost. Interviewers listen for which data belongs in a Server Component and which in a client cache, why a Server Action needs its own authorization check when the page rendering it is already protected, which environment variables a Next.js build copies into the browser bundle, and how a blocked event loop shows up in latency percentiles before it reaches an error log.
Node.js sits in the opening module with [JavaScript](/topics/lang-javascript) and [TypeScript](/topics/lang-typescript), far ahead of the frameworks that later run on it. The reason is [its event loop](/topics/be-nodejs-event-loop): browser and server schedule work on the same single-threaded model, Node adds phases and queues of its own, and the ordering questions of the first technical round draw on both. Algorithms follow for the coding round. The browser half builds a page from the markup up, through the styling tools to the [browser platform](/topics/fe-browser-platform) every framework sits on. The tooling before React is chosen for one repository holding both halves: Git, npm [workspaces](/topics/dev-npm-workspaces), where a types package shared by client and server usually lives, and Vite for the client build. From React onward the plan has a grammar: a concept module first, then the libraries that implement it. [UI framework concepts](/topics/fe-ui-frameworks-concepts) come before [React](/topics/fe-react) and its state libraries, [meta-framework concepts](/topics/fe-meta-frameworks-concepts) before [Next.js](/topics/fe-nextjs), React Router and TanStack Query. The concept answers for whichever library the interviewer's team chose; the library answers the follow-up. HTTP and the API styles, [REST](/topics/proto-rest) with GraphQL and OpenAPI beside it, sit between those pairs, because Next.js caching is argued in HTTP's terms of freshness and revalidation. The modules after Next.js ask what is wrong with a page that already works: whether it is fast, tested and accessible, and whether its tokens and forms are safe. [Frontend system design](/topics/fe-system-design), beside the general module, is where those answers meet, and the plan takes in delivery pipelines and coding with an AI assistant before it crosses to the server. The server half opens on CS theory and software design, then repeats the grammar. [Web framework concepts](/topics/be-web-frameworks-concepts) precede [Express](/topics/be-express), which leaves structure to you, and NestJS, which imposes modules, injection and the decorators TypeScript introduced at the start. [Zod](/topics/be-zod-parse-safeparse) shares their module because the request boundary is where it works: it parses what arrives at runtime, where TypeScript's types, erased at compile time, check nothing. Then the plan moves past the JSON endpoint: gRPC for contract-first calls between services, WebSockets and Server-Sent Events for connections a Node process keeps open long after an ordinary response would have ended. [ORM concepts](/topics/be-orm-concepts) precede Prisma and Mongoose, and the stores beneath them follow, relational and SQL first, then document modelling and [MongoDB](/topics/db-mongodb) itself, which defines the operators, indexes and transactions that Mongoose only expresses. Architecture and distributed systems turn the server half toward the senior design round. Security returns on the server, with [sessions, tokens and access control](/topics/be-authz) enforced where the data lives, and operations runs from Linux and containers to brokers and service levels. Only after all of that comes the JavaScript test tooling, apart from testing practice: runners and browser drivers, from [Jest](/topics/dev-jest) to Playwright, arrive when there is code on both sides of the stack to point them at. The [behavioral round](/topics/found-behavioral) and the interview process close the plan.
A React developer is closer to the server than the module order suggests: Server Components and Server Actions already run React code there, so the Next.js module is the bridge, and the Servers & Validation module, with the persistence and database modules below it, is the new ground. A Node developer's gap is narrower than it looks, since the language and the event-loop model are shared; what is new is React's render cycle, and effects above all, which behave unlike request handlers: they run after a render, run again when a dependency changes, and often owe a cleanup. Someone whose experience is one Express and React project gets most from the opening module and from SQL, which a document store let them skip. With a fortnight or less, the must know questions are the ones to finish: set importance there, and give the runtime, React and HTTP the hours first, since most rounds on this stack lean on them. Once both halves feel solid, switch study order to random, because the technical conversation on this stack jumps from a hook to a Node stream to a join, and practice should jump with it. Senior candidates should read frontend system design and the distributed-systems stretch side by side, since the design brief can come from either half. React is the only UI framework the plan names. If the posting asks for Vue, the concept modules still apply and the [Vue roadmap](/roadmaps/vue) supplies the framework; if the team also ships a mobile app from the same JavaScript, [React Native](/roadmaps/react-native) covers what the browser modules do not.
Two data centres hold the same data. The link between them breaks. Each must now answer from its own copy, or refuse. CAP is the proof that there is no third way.
A JavaScript full stack loop tests one language in two runtimes. The screen and the coding round are in JavaScript or TypeScript, and many teams swap the algorithm problem for a slice of the stack itself, a React view over an Express or NestJS endpoint, written live or set as a take-home. Technical conversations on the runtime, React and the server follow. From middle level up a design round is added, its brief as likely to be a feed or a typeahead as a backend service, and a behavioral interview closes.
Short snippets carry most of the technical conversation, and several of them only make sense with both runtimes in view. A zero-delay setTimeout and an already resolved promise: which callback runs first, and where does process.nextTick fall relative to both? Why does one synchronous loop over a large array inside a route handler stall every other request on that process? Why does an effect with an empty dependency array keep reading the count from its first render? A Date leaves the server as a Date and arrives in the browser as a string, while the type both sides import still promises a Date: what checks that payload, if the compiler does not?
A junior is expected to make the component and the endpoint work and to predict what those snippets do. Seniors are judged on placement: which process a line runs in, and what moving it would cost. Interviewers listen for which data belongs in a Server Component and which in a client cache, why a Server Action needs its own authorization check when the page rendering it is already protected, which environment variables a Next.js build copies into the browser bundle, and how a blocked event loop shows up in latency percentiles before it reaches an error log.
Node.js sits in the opening module with JavaScript and TypeScript, far ahead of the frameworks that later run on it. The reason is its event loop: browser and server schedule work on the same single-threaded model, Node adds phases and queues of its own, and the ordering questions of the first technical round draw on both. Algorithms follow for the coding round.
The browser half builds a page from the markup up, through the styling tools to the browser platform every framework sits on. The tooling before React is chosen for one repository holding both halves: Git, npm workspaces, where a types package shared by client and server usually lives, and Vite for the client build.
From React onward the plan has a grammar: a concept module first, then the libraries that implement it. UI framework concepts come before React and its state libraries, meta-framework concepts before Next.js, React Router and TanStack Query. The concept answers for whichever library the interviewer's team chose; the library answers the follow-up. HTTP and the API styles, REST with GraphQL and OpenAPI beside it, sit between those pairs, because Next.js caching is argued in HTTP's terms of freshness and revalidation. The modules after Next.js ask what is wrong with a page that already works: whether it is fast, tested and accessible, and whether its tokens and forms are safe. Frontend system design, beside the general module, is where those answers meet, and the plan takes in delivery pipelines and coding with an AI assistant before it crosses to the server.
The server half opens on CS theory and software design, then repeats the grammar. Web framework concepts precede Express, which leaves structure to you, and NestJS, which imposes modules, injection and the decorators TypeScript introduced at the start. Zod shares their module because the request boundary is where it works: it parses what arrives at runtime, where TypeScript's types, erased at compile time, check nothing. Then the plan moves past the JSON endpoint: gRPC for contract-first calls between services, WebSockets and Server-Sent Events for connections a Node process keeps open long after an ordinary response would have ended. ORM concepts precede Prisma and Mongoose, and the stores beneath them follow, relational and SQL first, then document modelling and MongoDB itself, which defines the operators, indexes and transactions that Mongoose only expresses.
Architecture and distributed systems turn the server half toward the senior design round. Security returns on the server, with sessions, tokens and access control enforced where the data lives, and operations runs from Linux and containers to brokers and service levels. Only after all of that comes the JavaScript test tooling, apart from testing practice: runners and browser drivers, from Jest to Playwright, arrive when there is code on both sides of the stack to point them at. The behavioral round and the interview process close the plan.
A React developer is closer to the server than the module order suggests: Server Components and Server Actions already run React code there, so the Next.js module is the bridge, and the Servers & Validation module, with the persistence and database modules below it, is the new ground. A Node developer's gap is narrower than it looks, since the language and the event-loop model are shared; what is new is React's render cycle, and effects above all, which behave unlike request handlers: they run after a render, run again when a dependency changes, and often owe a cleanup. Someone whose experience is one Express and React project gets most from the opening module and from SQL, which a document store let them skip.
With a fortnight or less, the must know questions are the ones to finish: set importance there, and give the runtime, React and HTTP the hours first, since most rounds on this stack lean on them. Once both halves feel solid, switch study order to random, because the technical conversation on this stack jumps from a hook to a Node stream to a join, and practice should jump with it. Senior candidates should read frontend system design and the distributed-systems stretch side by side, since the design brief can come from either half.
React is the only UI framework the plan names. If the posting asks for Vue, the concept modules still apply and the Vue roadmap supplies the framework; if the team also ships a mobile app from the same JavaScript, React Native covers what the browser modules do not.