Mastering API Retry Strategies: 8 Proven Methods for Reliable Backend Calls

In cross‑border services, third‑party servers are distributed worldwide, so network issues when calling external APIs are common; a retry mechanism becomes essential.

Retry Mechanism Implementation

1. Loop Retry

This is the simplest method: wrap the request in a loop and retry until success or the maximum attempts are reached.

Example code:

int retryTimes = 3;<br/>for(int i = 0; i < retryTimes; i++) {<br/>    try {<br/>        // request code<br/>        break;<br/>    } catch(Exception e) {<br/>        // handle exception<br/>        Thread.sleep(1000); // delay 1 second before retry<br/>    }<br/>}<br/>

The loop uses a for construct with a maximum of three attempts, and Thread.sleep() adds a delay to avoid rapid retries.

2. Recursive Retry

Recursion can also be used: the method calls itself on failure until success or the retry limit.

Example code:

public void requestWithRetry(int retryTimes) {<br/>    if(retryTimes <= 0) return;<br/>    try {<br/>        // request code<br/>    } catch(Exception e) {<br/>        Thread.sleep(1000); // delay 1 second<br/>        requestWithRetry(retryTimes - 1);<br/>    }<br/>}<br/>

3. Built‑in Client Retry

Many HTTP clients have built‑in retry support. For Apache HttpClient:

Version 4.5+:

HttpClients.custom().setRetryHandler(...)

CloseableHttpClient httpClient = HttpClients.custom()<br/>    .setRetryHandler(new DefaultHttpRequestRetryHandler(3, true))<br/>    .build();<br/>

Version 5.x:

HttpClients.custom().setRetryStrategy(...)

CloseableHttpClient httpClient = HttpClients.custom()<br/>    .setRetryStrategy(new DefaultHttpRequestRetryStrategy(3, NEG_ONE_SECOND))<br/>    .build();<br/>

Custom retry strategies can be implemented by providing HttpRequestRetryHandler (4.5+) or RetryStrategy (5.x).

4. Spring Retry Library

Spring Retry offers annotations and utilities for retrying methods.

<dependency><br/>    <groupId>org.springframework.retry</groupId><br/>    <artifactId>spring-retry</artifactId><br/>    <version>1.3.1</version><br/></dependency><br/>

Explicit RetryTemplate

Create a RetryTemplate and configure policies.

RetryTemplate retryTemplate = new RetryTemplate();<br/><br/>// configure retry policy<br/>RetryPolicy retryPolicy = new SimpleRetryPolicy(3);<br/>retryTemplate.setRetryPolicy(retryPolicy);<br/><br/>// configure back‑off policy<br/>FixedBackOffPolicy backOffPolicy = new FixedBackOffPolicy();<br/>backOffPolicy.setBackOffPeriod(1000);<br/>retryTemplate.setBackOffPolicy(backOffPolicy);<br/>

Execute the method via retryTemplate.execute(...).

retryTemplate.execute((RetryCallback<Void, Exception>) context -> {<br/>    // request code<br/>    return null;<br/>});<br/>

Alternatively, use @Retryable on methods:

@Retryable(value = Exception.class, maxAttempts = 3)<br/>public void request() {<br/>    // request code<br/>}<br/>

Annotation‑Based Retry

Enable retry with @EnableRetry and annotate methods with @Retryable, optionally configuring back‑off with @Backoff.

@Configuration<br/>@EnableRetry<br/>public class RetryConfig { }<br/><br/>@Service<br/>public class MyService {<br/>    @Retryable(value = MyException.class, maxAttempts = 3, backoff = @Backoff(delay = 1000))<br/>    public void doSomething() {<br/>        // request logic<br/>    }<br/>}<br/>

5. Resilience4j

Resilience4j provides lightweight retry, circuit‑breaker, and rate‑limiter features.

<dependency><br/>    <groupId>io.github.resilience4j</groupId><br/>    <artifactId>resilience4j-spring-boot2</artifactId><br/>    <version>1.7.0</version><br/></dependency><br/>

Programmatic Use

Create a RetryRegistry.

RetryRegistry retryRegistry = RetryRegistry.ofDefaults();<br/>

Configure a Retry instance.

RetryConfig config = RetryConfig.custom()<br/>    .maxAttempts(3)<br/>    .waitDuration(Duration.ofMillis(1000))<br/>    .retryOnResult(response -> response.getStatus() == 500)<br/>    .retryOnException(e -> e instanceof WebServiceException)<br/>    .retryExceptions(IOException.class, TimeoutException.class)<br/>    .ignoreExceptions(BusinessException.class, OtherBusinessException.class)<br/>    .failAfterMaxAttempts(true)<br/>    .build();<br/><br/>Retry retry = retryRegistry.retry("name", config);<br/>

Execute code with Retry.decorateCheckedSupplier.

CheckedFunction0<String> retryableSupplier = Retry.decorateCheckedSupplier(retry, () -> {<br/>    // request code<br/>    return "result";<br/>});<br/>

Annotation Use

@Service<br/>public class MyService {<br/>    @Retryable(value = MyException.class, maxAttempts = 3, backoff = @Backoff(delay = 1000))<br/>    public void doSomething() {<br/>        // request logic<br/>    }<br/>}<br/>

6. Custom Retry Utility

A lightweight custom retry framework can be built with a Callback abstract class, a RetryResult holder, and a RetryExecutor that loops until success or the retry limit.

public abstract class Callback {<br/>    public abstract RetryResult doProcess();<br/>}<br/><br/>public class RetryResult {<br/>    private Boolean isRetry;<br/>    private Object obj;<br/>    // constructors, getters omitted<br/>    public static RetryResult ofResult(Boolean isRetry, Object obj) {<br/>        return new RetryResult(isRetry, obj);<br/>    }<br/>    public static RetryResult ofResult(Boolean isRetry) {<br/>        return new RetryResult(isRetry, null);<br/>    }<br/>}<br/><br/>public class RetryExecutor {<br/>    public static Object execute(int retryCount, Callback callback) {<br/>        for(int cur = 0; cur < retryCount; cur++) {<br/>            RetryResult result = callback.doProcess();<br/>            if(result.isRetry()) continue;<br/>            return result.getObj();<br/>        }<br/>        return null;<br/>    }<br/>}<br/>

Usage example:

int maxRetryCount = 3;<br/>Object result = RetryExecutor.execute(maxRetryCount, new Callback() {<br/>    @Override<br/>    public RetryResult doProcess() {<br/>        // request logic<br/>        // return RetryResult.ofResult(true) to retry<br/>        // or RetryResult.ofResult(false, result) to finish<br/>    }<br/>});<br/>

7. Asynchronous Thread‑Pool Retry

Use a ThreadPoolExecutor to submit request tasks and retry asynchronously.

int maxRetryTimes = 3;<br/>int currentRetryTimes = 0;<br/><br/>ThreadPoolExecutor executor = new ThreadPoolExecutor(10, 10, 0L, TimeUnit.MILLISECONDS, new LinkedBlockingQueue<>());<br/><br/>Callable<String> task = () -> {<br/>    // request code<br/>    return "result";<br/>};<br/><br/>Future<String> future;<br/>while(currentRetryTimes < maxRetryTimes) {<br/>    try {<br/>        future = executor.submit(task);<br/>        String result = future.get();<br/>        break;<br/>    } catch(Exception e) {<br/>        currentRetryTimes++;<br/>        try { Thread.sleep(1000); } catch(InterruptedException ex) { Thread.currentThread().interrupt(); }<br/>    }<br/>}<br/>

8. Message‑Queue Retry

Persist failed requests to a message queue (e.g., RocketMQ) to ensure reliability.

@Component<br/>@RocketMQMessageListener(topic = "myTopic", consumerGroup = "myConsumerGroup")<br/>public class MyConsumer implements RocketMQListener<String> {<br/>    @Override<br/>    public void onMessage(String message) {<br/>        try {<br/>            // request code<br/>        } catch(Exception e) {<br/>            DefaultMQProducer producer = new DefaultMQProducer("myProducerGroup");<br/>            producer.setNamesrvAddr("127.0.0.1:9876");<br/>            try {<br/>                producer.start();<br/>                Message msg = new Message("myTopic", "myTag", message.getBytes());<br/>                producer.send(msg);<br/>            } catch(Exception ex) {<br/>                // handle send error<br/>            } finally {<br/>                producer.shutdown();<br/>            }<br/>        }<br/>    }<br/>}<br/>

Best Practices and Precautions

Set reasonable retry counts and intervals to avoid overwhelming the system.

Consider idempotency; avoid duplicate writes on non‑idempotent APIs.

Handle concurrency with locks or distributed locks to prevent duplicate requests.

Differentiate exception types: retry on transient network errors, handle permanent errors separately.

Avoid infinite loops; enforce maximum attempts to prevent resource exhaustion.