go-kata/04-errors-semantics/08-retry-backoff-policy

Kata 08: The Retry Policy That Respects Context

Target Idioms: Retry Classification, Error Wrapping (%w), Timer Reuse, Context Deadlines
Difficulty: 🟡 Intermediate

🧠 The "Why"

In other languages, retries are often hidden in SDKs. In Go, it’s easy to write:

• infinite retry loops,
• retry-on-any-error (bad),
• retry that ignores context cancellation (worse),
• retry implemented with repeated time.Sleep (hard to test, wasteful).

This kata makes you implement a testable, context-aware retry loop.

🎯 The Scenario

You call a flaky downstream service. You should retry only on transient failures:

• net.Error with Timeout() == true
• HTTP 429 / 503 (if you model HTTP)
• sentinel ErrTransient

Everything else must fail immediately.

🛠 The Challenge

Implement:

• type Retryer struct { ... }
• func (r *Retryer) Do(ctx context.Context, fn func(context.Context) error) error

1. Functional Requirements

• Retries up to MaxAttempts.
• Uses exponential backoff: base * 2^attempt with a max cap.
• Optional jitter (deterministic in tests).
• Stops immediately on ctx.Done().

2. The "Idiomatic" Constraints (Pass/Fail Criteria)

• Must NOT call time.Sleep inside the retry loop.
• Must use a time.Timer and Reset it (timer reuse).
• Must wrap the final error with context (attempt count) using %w.
• Must classify errors using errors.Is / errors.As.

🧪 Self-Correction (Test Yourself)

• If context cancellation only stops after the sleep: you failed.
• If you retry non-transient errors: you failed classification.
• If you can’t test it without real time: inject time/jitter sources.

📚 Resources

• https://go.dev/blog/go1.13-errors
• https://pkg.go.dev/errors
• https://pkg.go.dev/time