Rate Limiting & Throttling

Introduction

Rate limiting protects services from abuse and overload — token bucket, sliding window, and leaky bucket at gateway or service level.

The story

A partner API without rate limits got scraped; origin melted. Token bucket at gateway: 1000 req/min per API key — abusive key throttled, others unaffected.

The business problem

Teams that skip disciplined Rate Limiting & Throttling thinking pay in outages, cost overruns, and failed interviews:

• Outages: components chosen without scale math fail at peak.
• Cost: over-engineered microservices for 100-user products.
• Latency: missing cache/CDN/replica on read-heavy paths.
• Interviews: boxes without numbers and trade-offs don't hire.

The problem teams faced

This lesson addresses:

• When and why Rate Limiting & Throttling matters in real architectures.
• How to sketch components and data flows under interview time pressure.
• Trade-offs vs alternatives — what senior engineers articulate aloud.
• Production patterns and failure modes you've seen or will see.

Understanding the topic

Core idea: Rate Limiting & Throttling in production system design.

• Problem — what force this topic addresses.
• Building blocks — components typically involved.
• Trade-offs — what you gain and what you pay.
• Interview — how this appears in design loops.

Internal architecture

Rate Limiting & Throttling — system view:

Client → API Gateway
   ↓ rate limit check (Redis token bucket)
   ↓ allow / 429 Too Many Requests
Backend services protected
   ↓
Per-user, per-IP, per-API-key policies

Visual explanation

Three diagrams: architecture flow, design process, and scaling lens:

Informative example

Example — Rate Limiting & Throttling:

Token bucket: burst allowed, refill rate steady — good for APIs
Sliding window: precise count per minute — stricter
429 + Retry-After header; client exponential backoff

Execution workflow

Real-world use

Used in production at major tech companies and every FAANG system design interview loop. Patterns align with AWS/GCP well-architected frameworks and Google SRE practice.

Production case study

A partner API without rate limits got scraped; origin melted. Token bucket at gateway: 1000 req/min per API key — abusive key throttled, others unaffected.…

• Context: production or interview scenario from this lesson.
• Decision: component and trade-off choices explained.
• Outcome: measurable latency, availability, or cost impact.

Trade-offs

• Pro: structured approach reduces outages and interview failures.
• Con: upfront thinking takes time — faster than wrong rebuild.
• Con: every product has unique constraints — adapt templates.

Decision framework

• Always estimate before drawing microservices.
• Match consistency model to business domain (money vs likes).
• Document rejected alternatives — ADR or interview closing.

Best practices

• State assumptions explicitly (DAU, read:write ratio).
• Draw async vs sync paths with different line styles.
• Close with monitoring and on-call failure modes.

Anti-patterns to avoid

• Jumping to Kafka and microservices without scale justification.
• Single DB box with no read replica or cache on read-heavy design.
• Ignoring idempotency on write APIs with retries.

Common mistakes

• Underestimating peak QPS (forgetting peak factor).
• Hot shard from poor partition key choice.

Debugging tips

• Trace one request ID through diagram — find missing component.
• Ask "what fails first at 10×?" for every design.

Optimization strategies

• Cache + CDN first for read-heavy; shard when single DB saturates.
• Async queue for slow side effects (email, analytics).

Common misconceptions

• More components ≠ better design — simplest meeting NFRs wins.
• Interviews test process — not memorizing Netflix architecture.

Advanced interview questions

Summary

You can explain Rate Limiting & Throttling in a 45-minute system design interview with numbers, diagrams, and trade-offs. Teach it back without notes — you own this piece.