Reading — step 1 of 5
Read
~1 min readProduction
Putting It All Together
You've built every layer:
| Layer | Lesson |
|---|---|
| Why discovery | 1 |
| Registry | 2 |
| TTL + auto-deregister | 3 |
| Tags + metadata | 4 |
| Health checks | 5 |
| Load balancing | 6 |
| HA registries | 7 |
| Kubernetes | 8 |
| Service mesh | 9 |
Production patterns:
- Consul + service mesh: hybrid. Consul KV + Connect (mesh) is one stack.
- Kubernetes-native: Service object + headless services + maybe Istio.
- Eureka: still common in JVM-heavy shops.
- Custom registries: Netflix's history; bespoke for specific scaling needs.
What we glossed over:
- DNS-based discovery: SRV records carry port + weight + priority. Old but standardized.
- Pull vs push: registry can push updates to clients (websockets), or clients poll. Push reduces staleness.
- WAN federation: cross-DC discovery. Consul WAN gossip.
- External services: registering services that aren't part of your fleet (databases, third-party APIs).
For new infrastructure: use Kubernetes-native discovery + mesh (Istio/Linkerd) for advanced routing. Skip building your own — the existing options are mature.
Discussion
Ask a question, share an insight, or help someone who’s stuck.
Sign in to post a comment or reply.
Loading…