Course contentsShow
← System Design
Lesson 1028 of 1,91926. Idempotency and RetriesFree lesson

Client-Side vs Server-Side Retry Logic

Deciding where retry logic should live and understanding the tradeoffs of each approach for reliability.

Client-Side vs Server-Side Retry Logic

What you'll learn: Where to place retry logic in your architecture and the reliability tradeoffs of each location.

The Central Question

When a request fails in a distributed system, who should be responsible for retrying it—the client making the request, or the server receiving it? This seemingly simple question has significant implications for reliability, complexity, and failure modes.

Client-Side Retry Logic

The client implements exponential backoff, jitter, and retry attempts before giving up. This approach gives clients full control over their retry budget and timeout strategy.

Pros:

  • Clients can tailor retry logic to their specific needs (some may tolerate longer waits)
  • Network failures are naturally handled at the source
  • Reduces server complexity

Cons:

  • Every client must implement retry logic correctly (risk of inconsistency)
  • Clients may retry non-idempotent operations unsafely
  • Less visibility into retry storms from the server's perspective

Server-Side Retry Logic

The server accepts the request once, then internally retries failed operations (like downstream service calls or database writes) before responding.

Pros:

  • Centralized, consistent retry behavior
  • Can guarantee idempotency with idempotency tokens
  • Better observability of retry patterns
  • Shields clients from transient failures

Cons:

  • Client waits longer (blocking on server's retry loop)
  • Server holds resources during retries
  • If server crashes mid-retry, state may be lost

The Hybrid Reality

Most systems use both: clients retry network-level failures and timeouts, while servers retry internal operations like database queries or downstream API calls. Think of it as defense-in-depth—each layer handles failures at its level.

Key Takeaway: Client-side retries handle network failures and give flexibility; server-side retries ensure consistent, idempotent internal operations. Use both strategically based on where failures occur.