Distributed Systems

Aka “Microservices in the Cloud”.

Cloud

• Service Discovery.

• load balancing (resource utilisation).

• encryption?

• circuit breaker

  • Helps prevent cascading failures using a fast-fail service invocation when a service is misbehaving (just for a timeout period and then retried occasionally).
  • http://microservices.io/patterns/reliability/circuit-breaker.html

• saying “No” to distributed transactions and the alternatives.

• Failures. Partitions. Outages.

• CAP theorem.

Microservices

• Architecture

  • NGINX Microservices Reference Architecture (MRA) – 3 models: Proxy , Router Mesh, Fabric.

• Service discovery - local proxy or etcd/consul/zookeeper or Netflix/Spring-Cloud.

  • client-side – client communicates to Service Registry (e.g. etcd). e.g. Netflix OSS.
  • server-side – client communicates to a router (proxy/load-balancer) which in turn communicates with the Service Registry. e.g. Kubernetes proxy.
    • Kubernetes apparently has other ways of avoiding using the proxy pattern (which requires two networks hops per service invocation).

• Messaging/Queuing v REST (async v sync).

• JSON or binary serialisation.

• Service Invocation. Fan-out / Fan-in etc. Error handling. Reactive/Futures.

• Microservice Kit.

  • configuration, logging, health checks, metrics, distributed-tracing (aka correlation id)
  • examples:
    • Java
      • Dropwizard
      • Spring Boot / Spring Cloud
    • Go
      • Gizmo
      • Micro
      • Go kit

• Authentication

  • Access token e.g. JSON Web Token. https://jwt.io
  • Web app stateless authentication cookies (ala Rails / PlayFramework).

Persistence

• Databases - SQL, NoSQL, NewSQL.
• [Persistent] Queues / Messaging.
• Event streaming / CQRS / Kafka.