Idempotency for safe retries in HLD - System Design Guide

Problem Statement

When a client or network fails during a request, retrying the same operation can cause duplicate effects like double charges or repeated data entries. Without a way to detect and ignore repeated requests, the system can produce inconsistent or harmful results.

Solution

Idempotency ensures that repeating the same request multiple times has the same effect as doing it once. The system assigns a unique key to each request and stores the result. On retries with the same key, it returns the stored result instead of processing again, preventing duplicate side effects.

Architecture

Client

(sends req)

→API Gateway

↓

Idempotency Key

Store

This diagram shows the client sending a request with an idempotency key to the API Gateway, which checks the key store before forwarding to the service. The service processes the request and stores the result in the database. On retries, the gateway returns the stored result without reprocessing.

Trade-offs

✓ Pros

→

Prevents duplicate side effects from retries, ensuring data consistency.

→

Improves user experience by safely allowing automatic retries on failures.

→

Reduces load on backend services by avoiding repeated processing of the same request.

✗ Cons

→

Requires additional storage and management of idempotency keys and results.

→

Adds complexity to request handling logic and state management.

→

Needs careful expiration or cleanup of stored keys to avoid unbounded growth.

Use when operations have side effects that must not be repeated, especially in payment, order processing, or resource creation systems with retry-prone clients or unreliable networks.

Avoid when operations are naturally idempotent or stateless, or when retry rates are negligible and the overhead of managing keys outweighs benefits.

Real World Examples

Stripe

Stripe uses idempotency keys to ensure that payment requests retried due to network errors do not result in multiple charges.

Amazon

Amazon's order processing APIs use idempotency to prevent duplicate orders when customers retry submissions after failures.

Google Cloud Storage

Google Cloud Storage supports idempotent upload requests to avoid duplicate file writes on retries.

Alternatives

Retry with exponential backoff

Retries requests with increasing delays but does not prevent duplicate side effects if the operation is not idempotent.

Use when: Use when operations are safe to repeat or when idempotency is not feasible to implement.

Transaction-based processing

Uses database transactions to ensure atomicity but does not handle retries at the client or API level.

Use when: Use when the system can rely on strong database guarantees and retries are controlled internally.

Summary

Idempotency prevents duplicate effects from retrying the same request multiple times.

It works by storing and reusing results keyed by a unique request identifier.

This pattern is essential for operations with side effects in unreliable network environments.

Practice

1. What is the main purpose of idempotency in system design for safe retries?

easy

A. To increase the number of retries allowed

B. To ensure repeated requests produce the same result without side effects

C. To speed up the processing of new requests

D. To encrypt data during transmission

Solution

1. Step 1: Understand idempotency concept

   Idempotency means that making the same request multiple times results in the same outcome without causing additional changes.

2. Step 2: Identify the purpose in retries

   In safe retries, idempotency prevents duplicate processing and side effects by returning the same result for repeated requests.

3. Final Answer:

   To ensure repeated requests produce the same result without side effects -> Option B

4. Quick Check:

   Idempotency = Same result, no side effects [OK]

Hint: Idempotency means retrying won't change the outcome [OK]

Common Mistakes:

• Confusing idempotency with performance optimization
• Thinking idempotency increases retry limits
• Assuming idempotency encrypts data

2. Which of the following is the correct way to implement idempotency in an API request?

easy

A. Retry the request without any tracking

B. Encrypt the request payload for security

C. Ignore request IDs and process every request as new

D. Use a unique request ID to detect duplicates and avoid reprocessing

Solution

1. Step 1: Identify idempotency implementation method

   Idempotency requires tracking requests to detect duplicates, typically using a unique request ID.

2. Step 2: Evaluate options

   Only using a unique request ID allows the system to recognize repeated requests and avoid reprocessing.

3. Final Answer:

   Use a unique request ID to detect duplicates and avoid reprocessing -> Option D

4. Quick Check:

   Unique request ID = Idempotency key [OK]

Hint: Unique request ID prevents duplicate processing [OK]

Common Mistakes:

• Retrying blindly without tracking
• Ignoring request IDs
• Confusing encryption with idempotency

3. Consider a system that stores results of processed requests keyed by a unique request ID. If a client sends the same request twice with the same ID, what will the system return on the second request?

medium

A. A new result generated by reprocessing

B. An error indicating duplicate request

C. The original result from the first request

D. A timeout response

Solution

1. Step 1: Understand stored result usage

   The system caches the result of the first request using the unique request ID.

2. Step 2: Determine response for repeated request

   When the same request ID is received again, the system returns the cached original result to maintain idempotency.

3. Final Answer:

   The original result from the first request -> Option C

4. Quick Check:

   Repeated request returns cached result [OK]

Hint: Same ID returns cached result, not reprocessed output [OK]

Common Mistakes:

• Expecting an error on duplicate requests
• Assuming reprocessing happens every time
• Confusing timeout with duplicate handling

4. A developer implemented idempotency by storing request IDs but forgot to store the response results. What problem will this cause on retries?

medium

A. The system will reprocess the request causing side effects

B. The system will return the original result correctly

C. The system will reject all requests

D. The system will crash due to missing data

Solution

1. Step 1: Analyze missing response storage impact

   Without storing the response, the system cannot return the original result on retries.

2. Step 2: Identify retry behavior

   Since the system lacks the cached result, it reprocesses the request, causing potential side effects and breaking idempotency.

3. Final Answer:

   The system will reprocess the request causing side effects -> Option A

4. Quick Check:

   Missing response storage = Reprocessing on retry [OK]

Hint: Store both request ID and response to avoid reprocessing [OK]

Common Mistakes:

• Assuming storing request ID alone is enough
• Expecting system to reject retries automatically
• Thinking system crashes without response data

5. You design a payment system that must be idempotent to handle retries safely. Which combination of techniques best ensures idempotency and prevents duplicate charges?

hard

A. Use unique request IDs, store payment results, and check before processing

B. Allow retries without tracking, rely on client to avoid duplicates

C. Encrypt payment data and retry immediately on failure

D. Process every request independently to ensure payment completion

Solution

1. Step 1: Identify idempotency requirements for payments

   Payments must not be charged multiple times on retries, so tracking requests and results is essential.

2. Step 2: Evaluate techniques for safe retries

   Using unique request IDs to detect duplicates, storing results to return on retries, and checking before processing prevents duplicate charges.

3. Final Answer:

   Use unique request IDs, store payment results, and check before processing -> Option A

4. Quick Check:

   Unique ID + stored result = Safe payment retries [OK]

Hint: Combine unique IDs and result storage for safe retries [OK]

Common Mistakes:

• Relying on client to avoid duplicates
• Ignoring result storage
• Processing every request without checks