Service Discovery

আপনার backend-এ ১০টি instance of Order Service চলছে। Auto-scaling-এ সংখ্যা বদলায় — ৫ থেকে ২০। প্রতিটির আলাদা IP। User Service "Order Service" call করতে চায় — কোন IP-তে call করবে? এই dynamic problem-এর solution = Service Discovery।

Problem: Static Configuration ভাঙে

Traditional approach:

• Config file-এ "Order service IP: 10.0.0.5"।
• IP change হলে — config update + redeploy।
• Auto-scaling = constant change।
• Failed instance — request going there।

Service Discovery কী?

Service Discovery = একটি system যা track করে কোন service-এর কয়টি instance চলছে এবং প্রতিটির network location। Client এই service-কে query করে — current available instance পায়।

মূল components

• Service Registry: Database of all running services + locations।
• Service Provider: Service register itself।
• Service Consumer: Registry query করে location পায়।
• Health Check: Periodic check — alive কিনা।

Service Registration

Self-Registration

Service নিজেই startup-এ registry-তে register; shutdown-এ unregister; periodic heartbeat।

Third-party Registration

External agent (orchestrator) register/unregister করে। Kubernetes-এ kubelet automatic।

Discovery Patterns

Client-Side Discovery

Client registry query → instance list → load balance + call।

• সুবিধা: Smart routing client-এ।
• অসুবিধা: Client logic বেশি; multi-language complex।
• Tools: Netflix Eureka + Ribbon।

Server-Side Discovery

Client load balancer-কে call → LB registry query করে instance pick + forward।

• সুবিধা: Client simple।
• অসুবিধা: Extra hop (LB)।
• Tools: AWS ELB, Kubernetes Service।

জনপ্রিয় Tools

• Consul: HashiCorp — full-featured, DNS-based discovery।
• etcd: Kubernetes-এর backbone — distributed KV store।
• ZooKeeper: Apache, Hadoop ecosystem।
• Eureka (Netflix): AP-system, client-side discovery।
• Kubernetes DNS: Built-in — Service-এ ClusterIP।
• AWS Cloud Map: Managed service discovery।

DNS-Based Discovery

Service registration → DNS records। Client DNS lookup করে IP।

• Standard, simple।
• TTL caching issue।
• SRV records advanced (port info)।
• Kubernetes-এ default।

Kubernetes Service Discovery

K8s-এ service discovery built-in:

• Pod created → automatically registered।
• Service object — stable name (e.g., user-service)।
• kube-dns বা CoreDNS — DNS-based resolve।
• ClusterIP virtual — pod-এ load balance।
• Pod fail = automatically removed।

Health Check

Service alive কিনা detect:

• Heartbeat: Service periodically registry-তে ping।
• HTTP health endpoint: /health 200 return।
• TCP check: Port open কিনা।
• Custom script: Application-specific।

Failure detect → instance unregister।

Service Mesh — Modern Approach

Istio, Linkerd, Consul Connect — sidecar proxy:

• Service-to-service communication automate।
• Discovery + load balance + retry + circuit breaker সব।
• Application code-এ network logic নয়।
• Per-service Envoy proxy।

বাস্তব উদাহরণ

• Netflix: Eureka — pioneer।
• Airbnb: SmartStack (Consul-based)।
• Modern startup: Kubernetes built-in।
• HashiCorp ecosystem: Consul।

Challenges

• Eventual consistency: Registry update lag।
• Stale entries: Crashed instance still listed।
• Network partition: Registry unreachable = panic।
• Registry HA: Itself shouldn't be SPOF।

সাধারণ ভুল ধারণা

1. "DNS sufficient": Caching issue + no health check।
2. "Manual config OK at small scale": Auto-scaling-এ ভাঙে।
3. "Client-side > server-side": Use case-অনুযায়ী।

Best Practices

• Health check mandatory।
• Registry HA cluster (3-5 nodes)।
• Short TTL — quick failure detect।
• Service mesh-এ migrate consider — modern systems।
• K8s থাকলে — built-in use।
• Local cache + fallback if registry down।