Cloud-Native Applications

Introduction

Cloud-native applications are built from containers, deployed by Kubernetes, scaled by demand and observed via OpenTelemetry. MongoDB Atlas is the cloud-native data layer of choice: managed across AWS, Azure and GCP; provisioned via Terraform / Pulumi / Crossplane; observable via OTLP exporters.

This lesson shows the moving parts a cloud-native team wires together: Atlas IaC, workload identity for pods, PrivateLink networking, autoscaling and stream processing.

Understanding the topic

Cloud-native MongoDB stack:

• Atlas Terraform / Pulumi provider — clusters, users, network containers as code.
• Workload Identity / OIDC — pods authenticate to Atlas without static credentials.
• PrivateLink / VPC peering — cluster never touches the public internet.
• Atlas Auto-Scaling — compute & storage tiers grow without downtime.
• Atlas Search / Vector Search — managed, scaled with the cluster.
• Atlas Stream Processing — managed Kafka/MQTT/Kinesis stream pipelines.
• OpenTelemetry exporters — Atlas metrics → Prometheus / Datadog / Grafana.

Informative example

Atlas cluster as Terraform — versioned, reviewable, drift-detected:

resource "mongodbatlas_cluster" "prod" {
  project_id   = var.project_id
  name         = "prod-eu"
  cluster_type = "REPLICASET"
  provider_name               = "AWS"
  provider_region_name        = "EU_WEST_1"
  provider_instance_size_name = "M30"
  auto_scaling_disk_gb_enabled   = true
  auto_scaling_compute_enabled   = true
  provider_auto_scaling_compute_min_instance_size = "M30"
  provider_auto_scaling_compute_max_instance_size = "M60"
  backup_enabled               = true
  pit_enabled                  = true
}

Real-world use

Cloud-native fintechs and AI startups provision their entire data plane from a single Terraform module — Atlas cluster, IP allowlist, PrivateLink, database users, alerts — and ship platform changes as PRs reviewed like application code.

Best practices

• Everything as code — Atlas project, clusters, users, alerts, network — via Terraform.
• Use OIDC workload identity for pods; never mount static Atlas passwords.
• Disable public IP allowlist in production; PrivateLink only.
• Export Atlas metrics to your existing observability stack via OTLP.