InfraOps Self-Audit: A 20-Question Checklist for DevOps Careers

InfraOps Self-Audit: A 20-Question Checklist for DevOps Careers

Interviews for DevOps roles keep circling one theme: show me how you keep infrastructure honest. A self-audit is the fastest way to build a good answer, and it doubles as a useful exercise for your current team. This checklist walks through twenty questions, grouped into five areas, that you can answer in an afternoon with read-only access to a project. The structure follows the InfraOps write-up from the Ops Atlas, and I credit it whenever a point is specific. Grab a notebook, score each line yes, partly or no, and be harsh; nobody else needs to see the sheet.

Area 1: Visibility (can you say what exists?)

  • Can you list every account, region and resource type that is in scope, and what is deliberately excluded?
  • Do you know your coverage figure, meaning managed resources divided by in-scope resources?
  • Does anything run in production that no code describes? Name it.

If three or four of these come back as “no”, you have the classic gap that the Atlas highlights: a tidy change preview says nothing about resources it does not track. Start by listing the unmanaged items and ranking them by what hurts if they vanish.

Area 2: Change control (how does work reach production?)

  • Does every production change pass through a reviewed pipeline?
  • Has anyone applied a change from a personal laptop in the last month?
  • Is the change preview read by a second person before it is applied?
  • Do approvals sit on a protected environment rather than on a branch name?

The Atlas names laptop applies and skipped previews among the standard anti-patterns. It also warns that pull-request runs can execute a branch author’s edited workflow, so whatever role your preview job uses can be reached by anyone who can open a pull request. Treat that role as sensitive.

Area 3: State and structure (can you survive a mistake?)

  • Is any single state holding more than a few dozen resources?
  • Is the state locked during every apply, and does everyone know not to break a lock they did not take?
  • Are secrets kept out of state where the tooling allows it, using newer ephemeral and write-only features?
  • Are ignore rules documented, with a reason and a review date?
  • Are production databases protected both in code and at the provider?

Score yourself against the Google Cloud guidance quoted in the Atlas: well under a hundred resources per state, ideally a few dozen. For a career-minded angle on the thinking involved, our article on sharpening analytical habits for operations work pairs nicely with this section.

Area 4: Detection (will you hear about drift?)

  • Is a scheduled preview running against every root module, with a non-zero exit raising an alert?
  • Is each drift assigned to a named owner with a target date?
  • Do you record how long drift stays open?
  • Is there a written rule on when automatic reversal is acceptable, for instance limited to modules free of durable data, while nothing is being investigated?

The Atlas suggests roughly 72 hours as the alert-and-assign target for anything outside that narrow auto-fix category. Notice that the question is about the age of the oldest difference, not only the count.

Area 5: Rebuild and retirement (can you start over and clean up?)

  • Have you recreated a whole environment from code in the last six months?
  • Are machine images rebuilt on a schedule, with hosts replaced instead of patched in place?
  • Is there a process to remove retired environments, backed by code-derived inventory?
  • Does the team know which tests and policy checks run before merge?

A rebuild drill is the Atlas’s marker of its governed stage, and it tends to reveal forgotten console settings the first time around.

Scoring and next steps

Give a point for each yes and half a point for each partly. Under eight: you are essentially hand-built, so aim for the first 30 days of the Atlas plan, which cover a backend, plans in CI and one change shipped through the pipeline. Between eight and fourteen: you are codified, so add the review gate, a state split and nightly detection. Above fourteen: spend your effort on adopting the riskiest unmanaged items, rebuilding images weekly and passing a drill. The Atlas calls its maturity ladder a proposal, so treat the cut-offs as a rough guide rather than a verdict.

How to run this with a colleague

Doing the audit alone is useful; doing it in pairs is better. Ask a teammate to answer the same twenty lines independently, then compare. Every disagreement marks a spot where the documentation is thinner than anyone admitted, and those spots become your first tickets. Keep the session short, capture each “no” as a task with an owner, and schedule a repeat in a quarter. Resist fixing it all in a single sprint. Teams that attempt a big-bang clean-up usually stall, whereas two or three narrow changes per month, each with a measurable before and after, tend to stick.

Turning the audit into a career story

In interviews, describe the weakest area you found, what you changed and how you measured the change. A sentence such as “we went from unknown coverage to a number, and we shortened the oldest drift from weeks to days” lands better than a list of tool names. If you are considering a move toward cloud-focused roles, our guide on shifting from a DevOps role into cloud engineering covers the wider path. Every specific claim above can be traced to the Ops Atlas page on InfraOps, which is worth a slow read once you have your score.