slack-verify-1770279078/.claude/skills/microservices/SKILL.md

---
name: microservices
description: Inter-service communication patterns using pkg/svc for service discovery and circuit breaker protection. Use when implementing service-to-service calls.
---

# Microservices Communication

## Identity

You are a distributed systems engineer who understands the pitfalls of microservice communication. You prioritize resilience, observability, and graceful degradation over feature velocity.

## Service Discovery

Services discover siblings via environment variables injected automatically by the platform.

### How It Works

When a component is deployed, it receives env vars for all sibling services:
- `AUTH_SVC_URL=http://myproject-auth-svc:8001`
- `CHAT_SVC_URL=http://myproject-chat-svc:8002`

The naming convention: `{COMPONENT_NAME}_URL` where `COMPONENT_NAME` is UPPER_SNAKE_CASE.

### Using pkg/svc

```go
import "git.threesix.ai/jordan/slack-verify-1770279078/pkg/svc"

// Simple lookup
url := svc.ServiceURL("auth-svc")
if url == "" {
    // Service not configured
}

// Check availability
if svc.ServiceConfigured("auth-svc") {
    // Safe to call
}

// For required dependencies (panics if missing)
url := svc.MustServiceURL("auth-svc")
```

## Service Client

Use `svc.NewClient()` for a pre-configured HTTP client with circuit breaker protection.

### Basic Usage

```go
import "git.threesix.ai/jordan/slack-verify-1770279078/pkg/svc"

// Create client (returns error if service not configured)
authClient, err := svc.NewClient("auth-svc")
if err != nil {
    return fmt.Errorf("auth service unavailable: %w", err)
}

// Make requests
resp, err := authClient.Get(ctx, "/users/123")
if err != nil {
    if errors.Is(err, httpclient.ErrCircuitOpen) {
        // Circuit breaker is open - service is unhealthy
        return ErrAuthServiceDown
    }
    return fmt.Errorf("auth request failed: %w", err)
}
defer resp.Body.Close()

// JSON POST
resp, err := authClient.Post(ctx, "/validate", ValidateRequest{Token: token})
```

### Custom Configuration

```go
client, err := svc.NewClientWithConfig("auth-svc", svc.ClientConfig{
    Timeout:    5 * time.Second,  // Shorter timeout for fast-fail
    MaxRetries: 2,                // Fewer retries
    CircuitBreaker: &httpclient.CircuitBreakerConfig{
        FailureThreshold: 3,      // Open after 3 failures
        ResetTimeout:     15 * time.Second,
    },
})
```

## Circuit Breaker

The circuit breaker prevents cascading failures by failing fast when a service is unhealthy.

### States

| State | Behavior |
|-------|----------|
| **Closed** | Normal operation, requests pass through |
| **Open** | Blocks all requests, returns `ErrCircuitOpen` immediately |
| **Half-Open** | Allows one test request to check if service recovered |

### Default Thresholds

- Opens after 5 consecutive failures
- Waits 30s before attempting recovery (half-open)
- Closes after one successful request in half-open state

### What Affects Circuit State

The circuit breaker tracks **transient failures** only:

| Response | Affects Circuit? | Reason |
|----------|-----------------|--------|
| HTTP 2xx/3xx | ✅ RecordSuccess | Service is healthy |
| HTTP 5xx | ✅ RecordFailure | Server error - transient |
| HTTP 429 | ✅ RecordFailure | Rate limited - transient |
| HTTP 4xx (except 429) | ❌ No effect | Client error - not service's fault |
| Network error | ✅ RecordFailure | Connection failed |
| Context cancelled | ❌ No effect | User/caller initiated |
| Timeout | ✅ RecordFailure | Service too slow |

**Key insight:** 4xx responses (bad requests, not found, unauthorized) don't trip the circuit because they indicate a problem with the request, not the service. A service returning 400s is still "healthy" from a circuit breaker perspective.

### Handling Circuit Open

```go
resp, err := authClient.Get(ctx, "/users/123")
if errors.Is(err, httpclient.ErrCircuitOpen) {
    // Option 1: Return degraded response
    return CachedUserData(userID)

    // Option 2: Propagate as service unavailable
    return nil, ErrServiceTemporarilyUnavailable

    // Option 3: Use fallback service
    return fallbackClient.Get(ctx, "/users/123")
}
```

## Patterns

### Initialization Pattern

Initialize service clients at startup, not on-demand:

```go
type Server struct {
    authClient *svc.Client
    chatClient *svc.Client
}

func NewServer() (*Server, error) {
    authClient, err := svc.NewClient("auth-svc")
    if err != nil {
        return nil, fmt.Errorf("auth service required: %w", err)
    }

    // Optional dependency - check but don't fail
    var chatClient *svc.Client
    if svc.ServiceConfigured("chat-svc") {
        chatClient, _ = svc.NewClient("chat-svc")
    }

    return &Server{
        authClient: authClient,
        chatClient: chatClient,
    }, nil
}
```

### Response Decoding

```go
type User struct {
    ID   string `json:"id"`
    Name string `json:"name"`
}

resp, err := authClient.Get(ctx, "/users/123")
if err != nil {
    return nil, err
}

user, err := svc.DecodeResponse[User](resp)
if err != nil {
    return nil, fmt.Errorf("decode user: %w", err)
}
```

### Graceful Degradation

```go
func (s *Server) GetUserProfile(ctx context.Context, userID string) (*Profile, error) {
    // Required call
    user, err := s.fetchUser(ctx, userID)
    if err != nil {
        return nil, err
    }

    profile := &Profile{User: user}

    // Optional enrichment - don't fail if chat service is down
    if s.chatClient != nil {
        messages, err := s.fetchRecentMessages(ctx, userID)
        if err != nil {
            s.logger.Warn("failed to fetch messages", "error", err)
            // Continue without messages
        } else {
            profile.RecentMessages = messages
        }
    }

    return profile, nil
}
```

## Anti-Patterns

### Hardcoded URLs

```go
// BAD: Hardcoded URLs break when services move
client := httpclient.New(httpclient.Config{})
resp, err := client.Get(ctx, "http://auth-svc:8001/users")

// GOOD: Use service discovery
authClient, _ := svc.NewClient("auth-svc")
resp, err := authClient.Get(ctx, "/users")
```

### Ignoring Circuit Breaker Errors

```go
// BAD: Retrying forever when circuit is open
for {
    resp, err := authClient.Get(ctx, "/users")
    if err != nil {
        time.Sleep(time.Second)
        continue
    }
}

// GOOD: Detect circuit open and handle gracefully
resp, err := authClient.Get(ctx, "/users")
if errors.Is(err, httpclient.ErrCircuitOpen) {
    return nil, ErrServiceUnavailable
}
```

### On-Demand Client Creation

```go
// BAD: Creating client on every request
func (h *Handler) GetUser(w http.ResponseWriter, r *http.Request) {
    client, _ := svc.NewClient("auth-svc")  // Wastes resources
    // ...
}

// GOOD: Reuse client instance
type Handler struct {
    authClient *svc.Client
}

func (h *Handler) GetUser(w http.ResponseWriter, r *http.Request) {
    resp, _ := h.authClient.Get(r.Context(), "/users")
    // ...
}
```

### Silent Failures

```go
// BAD: Swallowing errors
resp, _ := authClient.Get(ctx, "/validate")
if resp != nil && resp.StatusCode == 200 {
    // Assume success
}

// GOOD: Explicit error handling
resp, err := authClient.Get(ctx, "/validate")
if err != nil {
    return fmt.Errorf("auth validation: %w", err)
}
if resp.StatusCode != http.StatusOK {
    return fmt.Errorf("auth validation failed: %d", resp.StatusCode)
}
```

## Checklist

When implementing inter-service calls:

- [ ] Use `svc.NewClient()` instead of raw HTTP clients
- [ ] Handle `ErrCircuitOpen` explicitly
- [ ] Initialize clients at startup, not on-demand
- [ ] Log service call failures with context
- [ ] Consider graceful degradation for optional dependencies
- [ ] Set appropriate timeouts (shorter than HTTP handler timeout)
- [ ] Propagate trace IDs for distributed tracing

## Files

| File | Purpose |
|------|---------|
| `pkg/svc/discovery.go` | Service URL lookup from env vars |
| `pkg/svc/client.go` | Pre-configured service client |
| `pkg/httpclient/circuit.go` | Circuit breaker implementation |
| `pkg/httpclient/client.go` | HTTP client with retries |