Deploy to Azure (Container Apps + Cosmos DB Mongo API)

The standard Azure shape is Container Apps for the app + Cosmos DB with the MongoDB API for the database. Container Apps gives you Kubernetes-shaped deployment with managed autoscaling but without the cluster overhead. Cosmos DB’s MongoDB API is wire- compatible with MongoDB (against a recent server version); behaviour is mostly the same but the cost / capacity model is Cosmos’s request-unit (RU) accounting rather than CPU/IOPS.

Quick reference


Cost (small app)	$30+/mo (Container Apps min + Cosmos Serverless or 400 RU/s)
Managed Mongo?	Cosmos DB (Mongo API) — or pair with Atlas
Cold start?	Scale-to-zero supported but ~5-10s wake
TLS	Automatic on `*.azurecontainerapps.io` + custom domains
HA	Multi-zone in one region; multi-region via Cosmos global distribution

Prerequisites

Azure subscription.
Azure CLI (az login).
An Azure Container Registry (or use Docker Hub).

Architecture

Internet → Container Apps Environment ingress (TLS)
             ↓
           Container App revision (1+ replicas, autoscaling)
             ↓
           Cosmos DB (Mongo API) — virtual network endpoint

Deploy

Resource group + container registry.

az group create --name davepi-rg --location eastus

az acr create --resource-group davepi-rg \
  --name davepiregistry --sku Basic --admin-enabled true

Build + push the image.

az acr login --name davepiregistry
docker buildx build --platform linux/amd64 --target runner \
  -t davepiregistry.azurecr.io/davepi:latest --push .

Provision Cosmos DB (Mongo API).

az cosmosdb create \
  --resource-group davepi-rg \
  --name davepi-cosmos \
  --kind MongoDB \
  --server-version 7.0 \
  --capabilities EnableServerless

Serverless is the cheapest starting tier — you pay per request unit. For predictable load, switch to provisioned throughput (400 RU/s minimum, ~$24/mo).

Note the connection string:

az cosmosdb keys list --resource-group davepi-rg \
  --name davepi-cosmos --type connection-strings \
  --query 'connectionStrings[0].connectionString' -o tsv

Container Apps environment.

az containerapp env create \
  --resource-group davepi-rg \
  --name davepi-env \
  --location eastus

Deploy the app.

az containerapp create \
  --resource-group davepi-rg \
  --name davepi \
  --environment davepi-env \
  --image davepiregistry.azurecr.io/davepi:latest \
  --target-port 4001 \
  --ingress external \
  --registry-server davepiregistry.azurecr.io \
  --secrets \
    mongo-uri="<the connection string from step 3>" \
    token-key="$(node -e 'console.log(require(\"crypto\").randomBytes(48).toString(\"hex\"))')" \
  --env-vars \
    NODE_ENV=production \
    PORT=4001 \
    MONGO_URI=secretref:mongo-uri \
    TOKEN_KEY=secretref:token-key \
    CORS_ORIGINS=https://your-frontend.example.com \
  --min-replicas 0 \
  --max-replicas 10

Verify.

curl "https://$(az containerapp show --resource-group davepi-rg \
  --name davepi --query 'properties.configuration.ingress.fqdn' -o tsv)/healthz"

Custom domain + TLS

Container Apps → custom domain → upload your TLS cert (or use Azure-managed certs). Configure the DNS validation records the portal lists.

Backups

Cosmos DB takes periodic backups automatically (default 4-hour intervals, 8-hour retention). For longer retention, switch the backup policy to “continuous” in the Cosmos account settings — that enables point-in-time restore up to 30 days.

For off-account retention, schedule a Container Apps Job that runs mongodump and writes to Azure Blob Storage.

Scaling

Container Apps: configure scaling rules per replica (HTTP request count, CPU, memory, custom). --min-replicas 0 is scale-to-zero; lift to 1+ to avoid cold starts.
Cosmos Serverless: capacity scales with request rate; no manual scaling needed. The trade-off is per-request cost vs predictable monthly cost.
Cosmos Provisioned: dial RU/s up/down. Auto-scale option scales between configured floor / ceiling.
Multi-region: enable Cosmos global distribution to replicate the DB to other Azure regions. Replicate Container Apps to matching regions for latency-sensitive users.

Observability

Logs: stdout → Log Analytics workspace (configured on the Container Apps environment). Kusto queries for ad-hoc searching; ship to Datadog / Splunk via Event Hubs.
Metrics: Container Apps built-in metrics (request count, latency, replica count). App-level /_metrics via the OTel recipe at Observability.
Tracing: Application Insights with the OTel collector sidecar.

Cosmos vs Atlas trade-off

Cosmos (Mongo API) is wire-compatible with MongoDB but not behaviourally identical:

✅ Global distribution + multi-region writes — one of Cosmos’s killer features. No equivalent in Atlas.
✅ Tight Azure integration (Managed Identity, VNet, Private Endpoints).
⚠️ Request-unit cost model. Easy to over- or under-provision; RU accounting takes some practice.
⚠️ Some MongoDB operators / aggregation stages aren’t supported, depending on the server version Cosmos exposes.

If your data shape leans on transactions, change streams across collections, or specific aggregation operators, validate against the current Cosmos Mongo API compatibility matrix before committing — or pair Container Apps with MongoDB Atlas (on Azure, via private endpoint) for real-MongoDB behaviour.