冪等性とリトライの仕組み
冪等なAPI操作を設計し、一時的な障害に副作用なく適切に対応できるインテリジェントなリトライ機構を実装します。
「冪等性とリトライの仕組み」はCoddyKit上の無料API Rate Limiting & Scalability Patternsレッスンです。 これはレッスン2/4です。 下記で完全なレッスンを無料で読むことができます。その後、ブラウザ内の組み込みコードエディタと24時間対応のAIチューターでハンズオン演習できます。 これはAPI Rate Limiting & Scalability Patterns学習パスの一部であり、ウェブとCoddyKitアプリ全体で進捗が同期されます。 API Rate Limiting & Scalability Patternsコースには全4レッスンが含まれています。
このレッスンの一部はまだ翻訳されておらず、英語で表示されています。
What is Idempotency?
In distributed systems, operations can sometimes fail or be interrupted. Idempotency is a property of an operation that means applying it multiple times produces the same result as applying it once.
Think of it like repeatedly pressing an 'On/Off' button. If it's truly idempotent, the first press changes the state, but subsequent presses (without an intervening 'Off') don't change it further. The end state remains the same.
Why Idempotency Matters
Idempotency is crucial for building robust and reliable APIs, especially when dealing with network issues or transient server errors.
- Prevents Duplicate Actions: If a request fails mid-way, and the client retries it, idempotency ensures the operation isn't performed twice.
- Ensures Data Consistency: Avoids creating duplicate records or incorrect state changes.
- Supports Retries: It's a foundational concept that allows clients to safely retry requests without unintended side effects.
Idempotent vs. Non-Idempotent
Let's look at common HTTP methods and their idempotency:
- GET: Always idempotent. Retrieving data multiple times doesn't change it.
- PUT: Idempotent. Updating an entire resource multiple times results in the same final state.
- DELETE: Idempotent. Deleting a resource multiple times has the same effect as deleting it once (it remains deleted).
- POST: Generally not idempotent. Creating a new resource multiple times usually creates multiple new resources.
The key is the result, not the action itself.
Implementing Idempotency Keys
For non-idempotent operations like POST (e.g., creating an order or processing a payment), we can introduce an idempotency key.
This is a unique identifier (often a UUID) generated by the client and sent with the request. The server then uses this key to detect and ignore duplicate requests within a certain time frame.
Server-Side Idempotency Check
Here's a conceptual look at how a server might handle an idempotency key. The server checks if the key has already been processed for that specific operation.
import java.util.Map;
import java.util.concurrent.ConcurrentHashMap;
public class PaymentProcessor {
private Map<String, Boolean> processedKeys = new ConcurrentHashMap<>();
public String processPayment(String idempotencyKey, double amount) {
if (processedKeys.containsKey(idempotencyKey)) {
System.out.println("Duplicate request for key: " + idempotencyKey + ". Returning previous result.");
return "Payment already processed for key " + idempotencyKey;
}
// Simulate payment processing
System.out.println("Processing payment of $" + amount + " with key: " + idempotencyKey);
try {
Thread.sleep(100); // Simulate work
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
}
processedKeys.put(idempotencyKey, true);
return "Payment successful for key " + idempotencyKey;
}
public static void main(String[] args) {
PaymentProcessor processor = new PaymentProcessor();
// First attempt with a key
System.out.println(processor.processPayment("uuid-123", 100.00));
// Retry with the same key (should be ignored)
System.out.println(processor.processPayment("uuid-123", 100.00));
// New request with a different key
System.out.println(processor.processPayment("uuid-456", 50.00));
}
}Introduction to Retries
Even with idempotent operations, requests can still fail due to temporary issues like network timeouts, server overload, or brief service outages. This is where retry mechanisms come in.
A retry mechanism automatically re-attempts a failed operation after a short delay. Its goal is to overcome transient (temporary) failures and improve the reliability of API calls.
Basic Retry Logic
The simplest retry mechanism involves a fixed number of retries with a constant delay between attempts. While straightforward, this can sometimes overwhelm a recovering service if many clients retry simultaneously.
public class SimpleRetry {
public static void makeApiCall() {
int maxRetries = 3;
int retryCount = 0;
long delayMillis = 1000; // 1 second
while (retryCount < maxRetries) {
try {
System.out.println("Attempt " + (retryCount + 1) + ": Making API call...");
// Simulate an API call that might fail
if (Math.random() > 0.6) { // 40% chance of success
System.out.println("API call successful!");
return; // Exit if successful
} else {
throw new RuntimeException("Simulated API failure.");
}
} catch (RuntimeException e) {
System.out.println("API call failed: " + e.getMessage());
retryCount++;
if (retryCount < maxRetries) {
try {
System.out.println("Retrying in " + delayMillis + "ms...");
Thread.sleep(delayMillis);
} catch (InterruptedException ie) {
Thread.currentThread().interrupt();
System.out.println("Retry interrupted.");
break;
}
}
}
}
System.out.println("All retry attempts failed.");
}
public static void main(String[] args) {
makeApiCall();
}
}Exponential Backoff
To avoid overwhelming services and to give them more time to recover, exponential backoff is a better strategy. It progressively increases the delay between retry attempts.
For example, delays could be 1s, 2s, 4s, 8s, etc. This reduces the load on a struggling service and spreads out retry attempts over time.
Backoff with Jitter
Even with exponential backoff, if many clients fail and retry at the exact same exponential intervals, they can still create a 'thundering herd' problem, all hitting the service at the same time.
Jitter adds a random component to the backoff delay. This helps to smooth out the retry attempts, distributing them more evenly and preventing synchronized bursts of traffic.
import java.util.Random;
public class ExponentialBackoffRetry {
private static final Random random = new Random();
public static void makeApiCallWithBackoff() {
int maxRetries = 5;
long baseDelay = 500; // milliseconds
long maxDelay = 16000; // cap the delay at 16 seconds
for (int retryCount = 0; retryCount < maxRetries; retryCount++) {
try {
System.out.println("Attempt " + (retryCount + 1) + ": Making API call...");
// Simulate an API call that might fail
if (Math.random() > 0.7) { // 30% chance of success
System.out.println("API call successful!");
return; // Exit if successful
} else {
throw new RuntimeException("Simulated API failure.");
}
} catch (RuntimeException e) {
System.out.println("API call failed: " + e.getMessage());
if (retryCount < maxRetries - 1) {
long delay = baseDelay * (long) Math.pow(2, retryCount);
delay = Math.min(delay, maxDelay);
// Add jitter: random value between 0 and delay
long jitteredDelay = random.nextInt((int) delay);
try {
System.out.println("Retrying in " + jitteredDelay + "ms (base: " + delay + ")...");
Thread.sleep(jitteredDelay);
} catch (InterruptedException ie) {
Thread.currentThread().interrupt();
System.out.println("Retry interrupted.");
break;
}
}
}
}
System.out.println("All retry attempts failed after " + maxRetries + " retries.");
}
public static void main(String[] args) {
makeApiCallWithBackoff();
}
}Idempotency & Retries Together
Idempotency and retry mechanisms are a powerful combination for building resilient distributed systems.
- Retries handle transient network or service failures, increasing the chance of an operation succeeding.
- Idempotency ensures that if a retry happens for an operation that actually succeeded (but the client didn't receive confirmation), no harmful duplicate side effects occur.
Together, they allow clients to make API calls with confidence, knowing that temporary issues won't lead to data corruption or incorrect states.
Check Your Understanding
Which of the following statements about idempotency and retry mechanisms are TRUE?
Recap: Robust APIs
In this lesson, we explored two critical concepts for building highly scalable and resilient APIs:
- Idempotency: Operations that produce the same result whether applied once or multiple times, crucial for preventing duplicate side effects.
- Retry Mechanisms: Strategies like exponential backoff with jitter that allow clients to gracefully handle transient failures by re-attempting requests with increasing, randomized delays.
By combining idempotency with intelligent retry logic, you can design API interactions that are robust, reliable, and tolerant of the unpredictable nature of distributed systems.
よくある質問
「冪等性とリトライの仕組み」レッスンは無料ですか？
はい。「冪等性とリトライの仕組み」の完全なテキストはこのウェブで無料で読めます。インタラクティブに演習し（組み込みコードエディタと24時間対応のAIチューター）、API Rate Limiting & Scalability Patternsコースの残りをアンロックするには、CoddyKit PROにアップグレードしてください。 API Rate Limiting & Scalability Patternsコースには全4レッスンが含まれています。
「冪等性とリトライの仕組み」で何を学びますか？
冪等なAPI操作を設計し、一時的な障害に副作用なく適切に対応できるインテリジェントなリトライ機構を実装します。 ブラウザで直接実行するハンズオンコードでAPI Rate Limiting & Scalability Patternsを演習し、24時間対応のAIチューターがレッスンを進める中での質問に答えます。
API Rate Limiting & Scalability Patternsを始めるのに経験は必要ですか？
事前経験は必要ありません。CoddyKitのAPI Rate Limiting & Scalability Patternsは初級者から上級者向けに構成されているため、ここから始めるか最初から始めて、自分のペースで進むことができます。 これはレッスン2/4です。
「冪等性とリトライの仕組み」レッスンにはどのくらい時間がかかりますか？
ほとんどのCoddyKitレッスンは約5～10分かかります。各レッスンはコンパクトでインタラクティブなので、着実に進歩し、ウェブとアプリ全体で正確に前回の場所から再開できます。
このAPI Rate Limiting & Scalability Patternsレッスンでコードを書いて実行できますか？
はい。すべてのAPI Rate Limiting & Scalability Patternsレッスンに組み込みコードエディタが含まれているため、ブラウザでリアルコードを書いて実行し、即座のAIフィードバックを取得できます。ローカル設定は不要です。