micro

name	micro
description	Microservices design and management.

Activate When

• /godmode:micro, "design microservices"
• "decompose this monolith", "service mesh"
• "saga pattern", "inter-service communication"

Workflow

1. System Context

find . -name "docker-compose*" -o -name "*.proto" \
  | head -10
grep -r "microservices\|grpc\|moleculer" \
  package.json 2>/dev/null

Architecture: Monolith | Modular Monolith | Microservices
Team Structure: <teams, ownership>
Scale: <request volume, data volume>
Data Stores: <databases, caches, brokers>

2. Service Decomposition

Domain-driven bounded contexts. Each service owns its data exclusively. Decompose by business capability (not technical layer).

IF two services need strong consistency: they belong together. IF greenfield: start with modular monolith.

3. Inter-Service Communication

• REST: CRUD, immediate response, external-facing. Circuit breaker (5 failures -> open), timeout 3s, retry 3x exponential, fallback.
• gRPC: high-throughput internal (10K+ RPS), low-latency, streaming, polyglot.
• Event-Driven: eventual consistency, decoupling, fan-out. Kafka (ordered), RabbitMQ (routing), SQS/SNS (AWS), NATS (ultra-low latency).

DECISION: Need immediate response -> sync (REST/gRPC). Fire-and-forget/fan-out -> async (events).

4. Service Mesh

• Istio: complex routing, multi-cluster, WASM. ~128Mi memory per sidecar.
• Linkerd: simpler, lighter (~64Mi), gentler learning curve.

5. Service Discovery

K8s DNS: <svc>.<ns>.svc.cluster.local. LB: Round Robin (homogeneous), Least Request (variable latency), Consistent Hash (sticky).

6. Saga Pattern

• Choreography (2-4 services): each publishes events, next reacts. Loose coupling, hard to trace.
• Orchestration (5+ services): central coordinator, compensates in reverse on failure. Easy to trace.

Every step MUST have compensating action (idempotent). Persist saga state. Use correlation IDs. Set timeouts.

7. Resilience Patterns

Circuit Breaker (5 consecutive 5xx, 30s half-open), Timeout (3s default), Retry (3x exponential+jitter), Bulkhead (10 concurrent per svc), Rate Limit, Fallback.

IF service boundary violation detected: flag for review. WHEN circular dependency found: split service.

8. Validation

Single bounded context per service, no shared DBs, single team ownership, async where possible, circuit breakers on sync calls, saga for distributed txns.

Quality Targets

• Target: <100ms inter-service call latency p95
• Target: >99.9% service availability
• Max service payload: <1MB per request
• Circuit breaker threshold: >50% error rate triggers open

Hard Rules

1. NEVER share a database between services.
2. NEVER decompose by technical layer.
3. NEVER skip modular monolith for greenfield.
4. EVERY sync call: circuit breaker + timeout 3s + retry.
5. EVERY service: health check + independent deploy.
6. NEVER use distributed transactions (2PC). Use sagas.
7. EVERY event: event_id, type, version, timestamp, source, correlation_id, data.

TSV Logging

Append .godmode/micro-results.tsv:

timestamp	services_count	communication	saga_pattern	mesh	status

Keep/Discard

KEEP if: contract tests pass AND circuit breakers work
  AND no cascading failures.
DISCARD if: contract broken OR cascading failure
  OR service discovery fails.

Stop Conditions

STOP when ALL of:
  - All boundaries defined with clear contracts
  - Circuit breakers on all inter-service calls
  - Contract tests passing in CI

Autonomous Operation

On failure: git reset --hard HEAD~1. Never pause.

Error Recovery

Failure	Action
Cascading failure	Add circuit breakers, bulkhead
Saga fails mid-way	Compensate reverse, verify idempotency
Discovery not resolving	Check DNS, retry with backoff
Circular dependency	Extract shared or use events