재시도 패턴 기초

Imagine you're trying to send a message, but your internet connection blips for a second. What do you do?

You probably try again! This simple human behavior is the core idea behind the Retry Pattern in software.

The Retry Pattern is a fundamental resilience technique. It involves automatically re-attempting an operation that has failed.

It's used when we expect the failure to be transient, meaning temporary and likely to resolve itself shortly, such as a brief network outage or a temporary database lock.

In distributed systems, services often depend on each other. Failures can occur for many reasons:

• Network issues: A brief disconnection or high latency.
• Resource contention: A database or service is temporarily overloaded.
• Service restarts: A dependent service is briefly unavailable during an update.

Retries help your application recover gracefully from these hiccups without crashing or requiring manual intervention.

At its simplest, the retry pattern works like this:

1. Attempt an operation.
2. If it fails, check if it's a retriable error.
3. If retriable, increment a counter and try again.
4. Stop after a certain number of attempts or if it succeeds.

Let's see a basic example without any delays yet.

This code simulates an operation that fails twice before succeeding. Notice how the while loop keeps trying until it works or runs out of attempts.

Simply retrying immediately might overwhelm a struggling service or fail again if the issue needs time to resolve. That's why we add delays.

A Fixed Delay Retry waits the same amount of time between each failed attempt. This gives the system a chance to recover.

Here, we've added a 1-second delay (1000ms) using Thread.sleep() after each failed attempt. This is a common practice to give the system some breathing room.

While fixed delays work, sometimes it's better to increase the wait time with each successive retry. This is called Exponential Backoff.

For example, you might wait 1s, then 2s, then 4s, then 8s. This reduces the load on a struggling service and gives it more time to recover.

Retries are most effective for:

• Transient network errors: Brief disconnections, timeouts.
• Temporary resource unavailability: A database connection pool is momentarily exhausted.
• Optimistic concurrency conflicts: When multiple users try to update the same record at once.
• Brief service restarts: A microservice is being redeployed.

Retries are not a silver bullet. Avoid using them for:

• Non-transient errors: Errors caused by invalid input, authorization failures, or missing resources that won't resolve on their own.
• Non-idempotent operations: If repeating an operation has unintended side effects (e.g., charging a customer twice). Idempotency means an operation can be performed multiple times without changing the result beyond the initial application.
• Long-lasting failures: If a service is permanently down or has a major outage.

Which scenario is generally a good candidate for applying the Retry Pattern?

You've learned the fundamentals of the Retry Pattern!

• It's for automatically re-attempting failed operations.
• It's crucial for handling transient failures in distributed systems.
• Basic implementation involves a loop with a maximum number of attempts.
• Adding delays (fixed or exponential backoff) is key to giving systems time to recover.
• Know when to use it (e.g., network issues) and when to avoid it (e.g., non-transient errors, non-idempotent operations).

Next, we'll explore other resilience patterns like fallbacks and timeouts!