Docs

Managed Postgres and Redis

POST /v1/databases provisions one engine per type: Postgres or Redis. There is no MySQL, no custom image, no HA.

{"name":"mydb","type":"postgres"}

type must be postgres or redis. Names are DNS-safe (alnum + hyphen, start/end alnum). Default quota: 3 databases per tenant (non-failed rows count).

Creation is a long route (no 30s chi timeout). Use an Idempotency-Key. Disk check 80/90 runs first.

Endpoints

Passwords and connection strings are AES-256-GCM encrypted at rest (tenant-derived key material via the master key).

Wiring a service

The platform does not auto-inject DATABASE_URL. You set env and restart:

CONN=$(curl -s -H "Authorization: Bearer $AH_KEY" \
  "$AH_URL/v1/databases/$DB_ID/connection-string" \
  | python3 -c 'import sys,json; print(json.load(sys.stdin)["connection_string"])')

curl -s -X POST "$AH_URL/v1/services/$SERVICE_ID/env" \
  -H "Authorization: Bearer $AH_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"DATABASE_URL\":\"$CONN\"}"

curl -s -X POST "$AH_URL/v1/services/$SERVICE_ID/restart" \
  -H "Authorization: Bearer $AH_KEY"

Restart recreates the container so new env vars take effect.

/ah-db and scripts/db-provision.sh do this pair of calls. ahc db is the CLI.

Runtime facts

• Containers use gVisor (runsc)
• Host bind is loopback-oriented in the manager; tenants reach DBs via the connection string the API returns
• Ports are unique (migration state_006)
• Readiness is protocol-level (Postgres/Redis), not “TCP accepted”
• Tenant suspend/delete stops database containers
• DELETE wipes volume data before removal
• GC looks for ah-db-* volumes not referenced in the DB (age > 10 minutes)

Stale provisioning rows are marked failed and cleaned on manager startup (ReconcileStale).

CLI

ahc db create --name mydb --type postgres
ahc db list
ahc db connection-string <id>
ahc db delete <id>