Scaling Patterns Trending

Backend engineers spent this week swapping concrete playbooks for taking an app from one machine to a fleet, with load balancers, hash-based routing, and read-heavy side systems at the center. (aws.amazon.com) A load balancer is the traffic cop in front of an app: Amazon says Elastic Load Balancing spreads requests across instances, containers, and IP addresses in multiple Availability Zones and sends traffic only to healthy targets. (docs.aws.amazon.com) The stateless part comes next. The Twelve-Factor App says apps should run as “one or more stateless processes,” and Kubernetes Deployments are built to manage workloads that “usually” do not maintain state. (12factor.net, kubernetes.io) Consistent hashing solves a different problem: keeping related requests on the same backend without reshuffling everything when servers are added or removed. Solo.io’s gateway docs describe it as soft session affinity based on headers, cookies, source IP, or query parameters. (docs.solo.io) That pattern shows up when teams need cluster routing, sticky caches, or shard-aware services. Fastly’s docs say consistent hashing can map requests to backends using request properties such as a path or user ID, while still redistributing traffic if one backend turns unhealthy. (fastly.com) The data side of the discussion split reads from writes. Microsoft’s Command Query Responsibility Segregation pattern says read and write operations can use separate models so each side is optimized independently for performance, scalability, and security. (learn.microsoft.com) In plain terms, that means transaction systems handle order placement or account updates, while a separate read system serves dashboards, search, or analytics. AWS Prescriptive Guidance says an API composer can call the microservices that own the data and combine the results with an in-memory join. (docs.aws.amazon.com) That is where polyglot microservices enters the picture: different services keep different kinds of databases, and the join moves out of one central schema and into an aggregation layer. Microservices.io lists both “Composition” for in-memory joins and “CQRS” for materialized read views as standard patterns for cross-service queries. (microservices.io) The production angle was less about theory than failure handling. Google Cloud’s load-balancing docs say health checks compute backend health from repeated probes, and failover guides route traffic to backup load balancers when outages are detected. (docs.cloud.google.com, docs.cloud.google.com) The through line in this week’s posts was simple: keep app servers disposable, move state into data stores and queues, and let routing, health checks, and separate read paths absorb growth. The result is not literally infinite capacity, but a system that can add machines without rewriting the whole app. (12factor.net, docs.aws.amazon.com, learn.microsoft.com)