Choosing compute and managed services

📘What it really is

• Compute choices, from most control to least work:
• virtual machines (EC2, Compute Engine, Azure VMs): you manage the operating system, patches and scaling;
• managed containers (ECS on Fargate, Cloud Run, Azure Container Apps): you give an image and settings; the provider runs and scales it. A good fit for RouteRani's API;
• managed Kubernetes (EKS, GKE, AKS): the most flexible for many services, with the most to learn (CD5);
• serverless functions (AWS Lambda, Cloud Run functions, Azure Functions): code that runs per event and costs nothing when idle; good for small, spiky tasks, with limits on run time and some start-up delay ("cold starts").
• Managed data services:
• databases (Amazon RDS and Aurora, Cloud SQL, Azure Database for PostgreSQL) handle backups, patching, standby copies and failover;
• object storage (S3, Cloud Storage, Azure Blob Storage) for files, backups and static websites, priced per gigabyte stored and transferred;

• queues and caches (SQS, Pub/Sub, ElastiCache/Memorystore).

• Choosing: prefer the most managed option that meets your needs, and spend your team's time on the product. Reasons to choose less managed: special software, strict control requirements, or cost at very large scale.

• Infrastructure as code for all of it: the same Terraform workflow as CD3, with the provider's resources, for example (not verified): hcl resource "aws_db_instance" "routerani" { identifier = "routerani-prod" engine = "postgres" instance_class = "db.t4g.medium" allocated_storage = 50 multi_az = true storage_encrypted = true username = "routerani" password = random_password.db.result deletion_protection = true }