---
title: Meta Backend Interview Questions (2026 Guide)
description: Real Meta backend engineer interview questions — the two-problem coding round, distributed system design, product architecture and the people round, with answers.
url: https://usegreenroom.app/blog/meta-backend-engineer-interview-questions
last_updated: 2026-08-01
---

← Back to blog

Global · Meta

# Meta backend interview questions and process

August 1, 2026 · 13 min read

![Meta backend interview questions guide — cover from Greenroom, the AI mock interviewer](/assets/blog/meta-backend-engineer-interview-questions-hero.webp)

"Design a system that tells me which of my friends are online." Easy, said the candidate, and drew a table: user ID, status, last seen. The interviewer nodded and asked how many rows. Three billion. Written how often? Every heartbeat. Read by whom? Everyone, constantly, on every page load. The table stayed on the whiteboard for another minute, looking increasingly like a small business plan for a very large fire.

That is the register of **Meta backend interview questions**. The prompts sound almost trivially simple — online status, a feed, a notification — and then the numbers arrive and every naive design collapses. Combine that with a coding round that fits two problems into 45 minutes and a behavioral people round that is genuinely scored, and you get a loop that rewards people who have thought about scale properly rather than memorised a template. I built Greenroom after walking out of an interview I had over-prepared for and under-performed in, so this guide covers what Meta asks and how to reason through it out loud.

## The Meta backend engineer interview process in 2026

- **Recruiter screen** — 20-30 minutes. Level calibration, logistics, and which product group you are being considered for.
- **Technical screen** — 45 minutes, typically two coding problems in an editor that does not execute your code.
- **Onsite (virtual, one day)** — usually two coding rounds, one system design round, and one behavioral people round. Senior candidates sometimes get a second design round in place of a coding one.
- **Hiring committee** — a packet review by people who never met you. The interviewer's written notes are the artefact, not the conversation.
- **Team matching** — the offer and the team are separate decisions. Expect a gap, sometimes weeks.

![Meta backend engineer interview process diagram — recruiter screen, technical screen, onsite with two coding rounds, system design and the people round, then hiring committee and team matching](/assets/blog/meta-backend-engineer-interview-questions-diagram.webp)

The Meta backend loop: four onsite rounds, a hiring committee that never met you, then team matching as a separate step.

Our [Meta interview preparation guide](/blog/meta-interview-preparation) covers the company-wide loop, and the [Meta frontend engineer guide](/blog/meta-frontend-engineer-interview-questions) covers the client side of the same products.

## The coding rounds: pacing beats difficulty

Most problems sit at LeetCode medium. The constraint is that there are two of them and 45 minutes total, in an editor with no compiler and no tests.

The rhythm that survives: two minutes clarifying, three minutes stating the approach and complexity aloud, twelve to fifteen minutes writing, two minutes tracing an example by hand. Then move — announcing "let me wrap this up so we have time for the second" is a plus, not an admission.

Graph traversal, intervals, heaps, hash-map counting, binary search on the answer, and string manipulation cover most of what gets asked. Meta interviewers tend not to say "that is wrong"; they ask a gentle question near the bug. Treat gentle questions as alarms.

```code:py
# interval merge shows up constantly, and the two-line invariant is the whole answer
def merge(intervals):
    intervals.sort(key=lambda x: x[0])        # sort by start, then sweep once
    out = []
    for start, end in intervals:
        if out and start <= out[-1][1]:       # overlaps the last kept interval
            out[-1][1] = max(out[-1][1], end) # extend it, do not append
        else:
            out.append([start, end])
    return out
```

Say the invariant while typing: "everything in `out` is disjoint and sorted, so I only ever need to compare against the last one." Our [DSA coding interview preparation guide](/blog/dsa-coding-interview-preparation-india) covers the problem list and [common coding interview mistakes](/blog/common-coding-interview-mistakes) covers the failure modes.

## System design at Meta scale

Meta's design prompts are deceptively small: online status, a news feed, a notification system, a comment thread, a URL shortener with Meta traffic. The skill being tested is whether you convert a product sentence into numbers before you draw anything.

- **Estimate first, out loud.** Users, requests per second, read-to-write ratio, payload size, storage per year. A design without numbers cannot be evaluated, and Meta interviewers will keep asking until you produce them.
- **Read-heavy or write-heavy.** The feed is read-heavy, so fan-out on write with a cache is defensible; celebrity accounts break it, so hybrid fan-out — precompute for normal accounts, fetch-on-read for the huge ones — is the answer they are listening for.
- **Cache strategy explicitly.** What is cached, where, with what invalidation, and what happens on a miss storm. Say "thundering herd" and explain request coalescing.
- **Sharding key with a failure story.** Any shard key you pick creates a hot shard somewhere; name the scenario and the mitigation. Our [database sharding interview questions](/blog/database-sharding-interview-questions) guide covers the tradeoffs.
- **Consistency choices per feature, not per system.** A like count can be eventually consistent; a payment cannot; a block list must be strongly consistent or you have a safety incident. Meta specifically likes this distinction.
- **Queues and back-pressure.** What happens when the consumer falls behind — drop, buffer, shed load, or degrade the feature? See our [Kafka interview questions](/blog/kafka-interview-questions) guide.
- **Failure and blast radius.** Which single component takes the product down, and what the degraded experience looks like rather than a blank screen.
- **Observability.** The three metrics you would alert on. Naming them unprompted separates senior answers from mid-level ones.

The online-status prompt is worth rehearsing specifically: heartbeats into an in-memory store with a TTL, fan-out only to people currently viewing you, accept a few seconds of staleness, and never write three billion rows to durable storage for data whose value expires in thirty seconds. The [Meta backend engineer prep page](/prep/meta-backend-engineer-interview) has a round-by-round checklist, and our [system design interview guide](/blog/system-design-interview-guide-india) covers the vocabulary.

## Backend fundamentals that come up as follow-ups

Meta rarely runs a separate fundamentals quiz, but the design round is full of them:

- **Databases** — index selection and why a query ignored your index, transaction isolation levels and the anomaly each one permits, and when a relational store stops being the right answer. See our [DBMS interview questions](/blog/dbms-interview-questions) guide.
- **Caching** — write-through versus write-back versus cache-aside, TTL selection, and stampede protection. Our [caching strategies interview questions](/blog/caching-strategies-interview-questions) guide covers it.
- **Concurrency** — races, idempotency keys, and what makes a retry safe. Retries without idempotency is the most common quiet bug in these designs.
- **Networking** — what actually happens across a request, connection reuse, timeouts and why a missing client timeout takes down a service.
- **APIs** — pagination, versioning and error contracts. Our [REST API interview questions](/blog/rest-api-interview-questions) guide covers the standard set.

## The people round: a real round, really scored

Meta's behavioral round is not a warm-down. It scores collaboration, conflict handling, ambiguity tolerance and honest self-assessment, and it breaks ties on the packet.

Bring five stories with real numbers: an outage you owned, a disagreement with a peer that you resolved, a project that slipped, something you drove without authority, and a piece of feedback that changed how you work. Rehearse each **out loud** to two follow-ups deep. Our [behavioral interview questions guide](/blog/behavioral-interview-questions-answers-software-engineer) and [tell me about a time you failed](/blog/tell-me-about-a-time-you-failed) guide cover the structure.

## LeetCode, Grokking, ChatGPT — where each fits

- **LeetCode** — Meta-tagged mediums, practised in pairs against a 45-minute clock so you train pacing, not just correctness.
- **Grokking the System Design Interview / Designing Data-Intensive Applications** — the second is genuinely the best single book for the design round's vocabulary and for the consistency questions Meta asks.
- **Meta's engineering blog** — their public writing on TAO, memcache at scale and feed ranking describes the exact problems the design round abstracts.
- **ChatGPT** — good for generating prompts and reviewing a written design. It will not ask "how many rows?" at the moment your table has three billion of them.
- **Greenroom** — the spoken layer. [Ari, the AI interviewer](/) runs the design round out loud, pushes the second and third follow-up, and scores structure and clarity. Honest tradeoff: Ari will not audit your code line by line, so pair it with LeetCode.

**The core truth:** Meta's design round is not testing whether you know the components. It is testing whether you reach for numbers before you reach for boxes — and that habit only forms when you have been asked "how many?" out loud, repeatedly, by someone who is not you.

## How to prepare for the Meta backend interview

- **Week 1** — Meta-tagged mediums in pairs under a 45-minute clock. End every problem by stating complexity aloud.
- **Week 2** — estimation drills. Take five product sentences and produce users, QPS, read/write ratio and yearly storage for each in under three minutes, spoken.
- **Week 3** — design the feed, online status, notifications and a rate limiter out loud, numbers first, with a named failure story for each.
- **Final week** — five people-round stories with numbers rehearsed to two follow-ups deep, two full spoken mocks, and the [role-specific prep page](/prep/meta-backend-engineer-interview) the night before. Nothing new in the last 48 hours.

Interviewing across comparable loops? The [Google backend engineer guide](/blog/google-backend-engineer-interview-questions) covers a similar bar, the [Uber backend engineer guide](/blog/uber-backend-engineer-interview-questions) covers dispatch-scale systems, and the [Flipkart backend engineer guide](/blog/flipkart-backend-engineer-interview-questions) covers the same problems at India-scale traffic spikes.

## Frequently asked questions

### What is the Meta backend engineer interview process?

Candidates report a recruiter screen, a 45-minute technical screen with typically two coding problems, and a virtual onsite of about four rounds — two coding, one system design, and a behavioral people round, with senior candidates sometimes getting a second design round. A hiring committee then reviews the written packet, and team matching happens separately after the hiring decision.

### What system design questions does Meta ask backend engineers?

Prompts are short and product-shaped: design the news feed, an online status system, notifications, a comment thread or a rate limiter. What is graded is whether you convert the sentence into numbers first — users, requests per second, read-to-write ratio, storage per year — then choose fan-out strategy, cache and invalidation, a shard key with a named hot-shard failure story, per-feature consistency, and back-pressure when a consumer falls behind.

### How many coding questions does Meta ask in one interview?

Two problems in 45 minutes is the standard shape, in an editor that typically does not run your code. Budget roughly twenty minutes per problem including clarification and a spoken trace of an example, and announce the move to the second problem rather than silently running out of time.

### How do you answer the Meta online status design question?

Do not store a durable row per user per heartbeat. Write heartbeats into an in-memory store with a short TTL, treat absence of a recent heartbeat as offline, fan out changes only to people currently viewing that user, and explicitly accept a few seconds of staleness. Then discuss the read amplification of a friend list, cache invalidation, and what the feature degrades to when the store is unavailable.

### What is the people round at Meta?

It is Meta's scored behavioral round covering collaboration, conflict, ambiguity and honest self-assessment, and it frequently breaks ties in the packet review. Prepare five real stories with numbers — an outage you owned, a peer disagreement you resolved, a project that slipped, work you drove without authority, and feedback that changed how you work — rehearsed out loud to two follow-ups deep.

### Is the Meta backend interview hard?

It is a high bar, but the difficulty is in shape rather than obscurity: the coding problems are mostly LeetCode medium under an unusually tight clock, and the design prompts are one sentence long but scale to billions of users the moment you ask for numbers. Candidates who have only read about system design, rather than talked through it, tend to stall at the second follow-up.

Meta's backend loop is decided by the follow-up question you did not expect, in the design round and the people round alike. [Greenroom](https://usegreenroom.app/) runs mock backend interviews out loud with Ari, pushes those follow-ups, and scores structure, clarity and depth. Free to start. Curious how it works? See [how AI mock interviews work](/blog/ai-mock-interview).