Host ai workflow runner with Starlette: Postgres-backed state

Summary

Deploy an ai workflow runner built with Starlette on Ample using the durable job state pattern. Compute runs the app in an isolated microVM behind a public HTTPS URL; a managed PostgreSQL 16 database is auto-provisioned and injected as DATABASE_URL. Verified on Starlette: a jobs table with a state and attempts counter, processing inside a row-locked transaction, and a second run that is a no-op (job=done attempts=1 idempotent=true). Not separately tested: your job types, payload schema and retry policy; treat the ai workflow runner-specific behavior as your application code.

Infrastructure requirements

• Compute: verified (Apps run in isolated x86_64 Firecracker microVMs that auto-pause when idle and wake on request; sizes are the priced VM sizes.)
• Postgres: verified (Managed PostgreSQL 16 runs in its own microVM and is auto-provisioned when an app needs a database and no DATABASE_URL is supplied.)

Prerequisites

• A Starlette project that builds and starts with the documented commands (pip install into .ample/python from requirements.txt, then uvicorn from run.py reading PORT on the python-3.12 template)
• A PostgreSQL driver reading DATABASE_URL at runtime (auto-provisioned when omitted, or supplied with --env)
• An Ample account token with servers:write, databases:read

Exact tested configuration

• template: python-3.12
• runtime: python
• size: s-1vcpu-1gb
• install: python3 -m pip install --target .ample/python -r requirements.txt
• start: PYTHONPATH=.ample/python:${PYTHONPATH:-} python3 run.py

Steps

1. Build and start. pip install into .ample/python from requirements.txt, then uvicorn from run.py reading PORT on the python-3.12 template. The server must bind 0.0.0.0 on PORT.
2. Implement the pattern on PostgreSQL. The fixture's module implements durable job state: a jobs table with a state and attempts counter, processing inside a row-locked transaction, and a second run that is a no-op (job=done attempts=1 idempotent=true). Copy the approach into your schema; keep migrations idempotent and run them with --release-command.
3. Deploy. Run the synchronous deploy once and read the result (exit 0 live, 1 failed, 2 blocked). Re-running with no change is a no-op.

   ample deploy . --name <app-name> --public --start "python3 run.py"

4. Verify. Fetch the live URL and the pattern self-test route(s) (/p/durable-job-state) from the example; then run your own checks. On failure read ample logs <deployment_id> --kind build then --kind runtime.

   ample logs <deployment_id> --kind build

Tested examples

• Starlette pattern fixture (tests/deploy-canaries/starlette-patterns): Multi-pattern Starlette app whose durable job state module was checked live; the module is under tests/deploy-canaries/_pattern-modules.

Success checks

• app responds on its public URL (/ on the live URL, expect ample canary starlette patterns)
• durable-job-state self-test (/p/durable-job-state on the live URL, expect job=done attempts=1 idempotent=true)

Limitations

• Verified on the python-3.12 template at s-1vcpu-1gb with the example fixture; other sizes, templates and Starlette major versions are not verified.
• The ai workflow runner itself (your job types, payload schema and retry policy) is application code and was not separately tested; the pattern checks are what was verified.
• Region, compliance attestations and request-duration limits are unknown and not claimed.
• Apps auto-pause when idle and wake on the next request; always-on is an operator setting, not a plan feature.
• Managed PostgreSQL 16 only; extensions, connection limits and backup or restore procedures are not verified; apps and their databases are placed together.

Cost estimate

Estimated 10.00 USD per month (size prices from pricing.toml at build revision a1b8c38919e59cd035ebabaced73cf84ece24371).

• app server x1 s-1vcpu-1gb: 5.00 USD
• managed PostgreSQL database x1 s-1vcpu-1gb: 5.00 USD

Always-on monthly price of the tested sizes; apps and databases auto-pause when idle. Buckets are allocation-priced per quota and not included.

Verification evidence

• canary_run on 2026-09-21T01:21:36Z at revision 50dbac567d2c5cc8e55e6c748a646be0025d9cfb-dirty (CLI 0.1.21): Starlette pattern fixture deployed on Ample (pip install into .ample/python from requirements.txt, then uvicorn from run.py reading PORT on the python-3.12 template); checks passed for durable-job-state and configuration-secrets. Pattern proof: a jobs table with a state and attempts counter, processing inside a row-locked transaction, and a second run that is a no-op (job=done attempts=1 idempotent=true). (expires 2027-03-20T01:21:36Z)
• canary_run on 2026-09-20T02:16:51Z at revision 199ff1dfd52683832ae75d3f98b53a7a4bff7f96-dirty (CLI e3181f5): The same Starlette app deployed with a --release-command migration; the marker it created was readable after activation. (expires 2027-03-19T02:16:51Z)

Last verified: 2026-09-21T01:21:36Z

Execution binding

MCP tool ample_deploy (registry mcp:ample_deploy), schema hash 876465fce906da0c observed 2026-09-21T03:19:59.461917+00:00 at revision 49962bcade4f, binding state current, required scopes: servers:write, databases:read.