SMS Orchestrator — Deployment Topology

Status: populated Owner: SRE Last updated: 2026-04-18

1. Runtime

• Node.js 22 LTS, TypeScript 5.x strict.
• NestJS 10 + @nestjs/platform-fastify.
• Image: distroless Node, multi-stage build, non-root.
• Listen ports: 3010 HTTP, 9464 metrics, 9100 gRPC (if future).

2. Kubernetes

Object	Value
Kind	Deployment
Replicas	3 (min) / HPA 3–10
HPA triggers	CPU 70%, custom metric orch_submit_inflight (target 60 per pod)
Rolling update	maxSurge=25%, maxUnavailable=0
PDB	minAvailable=2
Resources	requests: 500m CPU / 512Mi; limits: 2 CPU / 1Gi
ServiceAccount	sms-orchestrator (Vault K8s auth)

3. Dependencies (at runtime)

Dep	Mode	Notes
PostgreSQL	Primary writer + read replicas	Connection pool 20 per pod
Redis	Cluster (6 nodes)	Dedicated namespace; pipeline commands
NATS JetStream	3-node cluster	Durable consumer orch-consumer
routing-engine	gRPC	Client-side LB, timeout 200 ms, retry 1
Vault	Sidecar	JWT signing keys fetch at startup

4. Kong routes (upstream of this service)

See api-gateway/API_CONTRACTS. Relevant routes:

• /v1/sms/send → sms-orchestrator:3010
• /v1/sms/bulk → sms-orchestrator:3010
• /v1/sms/{id} → sms-orchestrator:3010

5. Regions

Current: single primary region, optional DR (assumption A-001 in 01). Active-active multi-region is a future work item.

6. Scaling model

• Submit path: CPU-bound on JSON + Zod; scales linearly.
• Pipeline path: mostly I/O wait; limited by routing-engine + NATS publish throughput.
• Benchmarked: 1500 submits/sec per pod at 70% CPU (4 vCPU).

7. Rollout

• Canary: 5% traffic for 30 min, 2 replicas.
• Promote if SLI-ok (error rate < 1%, latency P95 < 250 ms).
• Rollback = revert Deployment image tag; Kong needs no change.