Multi-Tenant Infrastructure Automation Platform

Shared cluster model balances cost efficiency with tenant isolation: all clients run on the same ECS cluster and share ALB infrastructure, reducing per-client overhead by 60-70%. Service-level isolation ensures each client has a dedicated ECS service with separate task definitions, scaling policies, and target groups for workload and resource isolation. Configuration-based tenancy uses a single application codebase serving all clients dynamically based on API-provided configurations, eliminating code duplication and simplifying updates. Network isolation via ALB listener rules and target groups prevents cross-client traffic.

EC2 Launch Type provides cost-effective compute for predictable, steady-state workloads with full control over instance types and scaling policies. Cluster Auto Scaling monitors aggregate CPU and memory reservation across all services and scales EC2 instances up or down to maintain target utilization (~70-80%). Service Auto Scaling adjusts the number of ECS tasks per client based on request rate, CPU, or memory metrics, ensuring responsive performance during traffic spikes.

Programmatic AWS API calls (boto3) from the FastAPI backend provision all infrastructure components: ECS services, ECR repositories, CodePipeline, CodeBuild projects, CodeDeploy applications, target groups, listener rules, Route 53 records. Idempotent operations ensure re-running provisioning for an existing client updates resources rather than failing, supporting infrastructure updates and drift correction. Infrastructure state tracking in PostgreSQL database records provisioning status, resource ARNs/IDs, and timestamps for auditing and lifecycle management. Error handling and rollback capabilities log failures and support manual or automated cleanup.

Single-repository strategy allows all clients to deploy from the same GitHub repository, simplifying codebase management and enabling simultaneous feature rollouts to all clients. Per-client pipelines ensure each client has a dedicated CodePipeline instance to isolate build artifacts, deployment history, and permissions. Blue-green deployment with CodeDeploy provisions new ECS tasks (green) alongside existing tasks (blue), performs health checks, and shifts traffic gradually; old tasks are terminated only after successful cutover, enabling instant rollback on failure. Build caching and optimization through CodeBuild using Docker layer caching and dependency caching (pip for Python, npm for Vue.js) to accelerate build times.

Centralized configuration API in the management portal exposes tenant-specific settings via authenticated endpoints. Runtime configuration fetching allows ECS tasks to call the configuration API on startup and cache settings locally, with support for hot-reloading on configuration updates. Secrets management stores sensitive client credentials (database passwords, API keys) in AWS Secrets Manager or SSM Parameter Store, retrieved by ECS tasks via IAM task roles. Environment variables pass non-sensitive configuration (client ID, API endpoint URLs) to ECS tasks via task definition.

Shared infrastructure (ECS cluster, ALB) amortizes fixed costs across all clients, reducing per-client infrastructure overhead by approximately 60-70% compared to dedicated clusters. EC2 auto-scaling ensures cluster capacity matches actual demand, avoiding over-provisioning and reducing idle compute costs. ECR lifecycle policies automatically delete old Docker images after 30 days or keep only the latest 10 versions, minimizing storage costs. Consolidated billing and cost allocation tags on all provisioned resources enable accurate per-client cost tracking and chargeback.