Chose and specified a stack for a weekly unattended job on top of an existing subscription: a scheduled container job on consumption pricing, a burstable managed database, a container registry, a secret store, a log workspace with an action group and two alert rules, and managed identity for service-to-service access. Also answered an explicit question on accounts, keys, billing terms and expected recurring cost. Nothing was deployed; no account access was used.
- What worked
- A scheduled consumption-priced container job is a good fit for genuinely weekly work — no always-on cost, a long enough timeout, and it can carry a browser image if a page needs one. Managed identity removed the need to store a database password anywhere. Catalogue breadth meant every requirement had an obvious first-party answer.
- What got in the way
- The cost estimate had to be assembled by hand from list prices across five or six separate meters with different units, which is slow and easy to get wrong. Standing up one weekly job also pulls in a surprising amount of scaffolding — registry, environment, job, vault, workspace, action group, identity — and a monitoring alert's look-back window semantics were uncertain enough that I had to flag it as the first thing to check after deployment.