The 2 a.m. Certificate Expiry and the Audit Checklist It Inspired

The 2 a.m. Certificate Expiry and the Audit Checklist It Inspired

At two in the morning on a quiet Saturday, an on-call engineer named Wei opened her laptop to a wall of red. An internal service certificate had expired. Every service that talked to it was failing, and nobody could say who owned the thing. The fix took four minutes once they found the right person. Finding that person took three hours. In the retrospective, someone dug up the certificate’s issue date and quietly pointed out that its expiry had been knowable for a year. That night Wei’s team started what this post is about: an expiry and renewal audit that feeds a single key-dates calendar. Below is the checklist they used, in the order they used it.

Why the dates are the real incident

When people say “the certificate expired,” the root cause is almost never cryptography. It is that a fixed, public, machine-readable date was not attached to a human being with a reminder. Infrastructure is full of such dates: certificates, licences, support windows, API deprecations, cloud commitments and domain registrations. They are not surprises. They are tickets nobody opened.

An audit turns that around. You find every dated thing, give each an owner and a start reminder, and put them all in one place. Doing it once is a project. Keeping it current is a habit of about fifteen minutes a quarter.

Step 1: Inventory everything with an end date

Start with discovery, because you cannot calendar what you have not found. Walk each of these areas and write down what you find.

  • Certificates: public web certificates, internal service and mutual-TLS certificates, code-signing certificates and any certificates embedded in appliances or devices.
  • Credentials with lifetimes: API keys, service-account keys and tokens that expire, plus cloud access credentials set to rotate.
  • Domains and DNS: registration expiry, nameserver provider contracts and any delegated sub-zones owned by other teams.
  • Licences and subscriptions: monitoring, CI/CD, security scanners, container registries, IDE seats and paid support contracts.
  • Cloud commitments: reserved capacity, savings plans and committed-spend terms that end on a date.
  • Support lifecycles: operating-system releases, Kubernetes minor versions, language runtimes and database engines, each with a published end-of-support date.

Expect to find things that nobody remembers owning. Treat each as a finding rather than a failing. A discovery sweep is supposed to turn up orphans.

Step 2: Record five fields for every item

Wei’s team agreed on a minimum record so entries stayed comparable. For each dated item, capture:

  1. What it is, in words a new colleague would understand.
  2. The exact expiry or deadline date.
  3. The owner and a named deputy.
  4. The renewal lead time, meaning how long the process really takes, including approvals and procurement.
  5. A link to the place where renewal happens, such as a vendor console or a runbook.

The lead time field matters most. A certificate that renews in seconds through automation and a licence that needs purchase-order approval from finance live on very different clocks, and the calendar must reflect that.

Step 3: Calculate the start date

Subtract the lead time from the deadline, then add a buffer for holidays and absences. The result is the date your first reminder fires. If a procurement cycle takes six weeks, your reminder fires at least two months out. Do not use a single default such as “thirty days” for everything. The difference between a good audit and a bad one is how honestly lead times are estimated.

Tie this back to the way you plan other work. The checklist at the heart of a cloud services checklist for growing without stress follows the same spirit: write down the dull recurring tasks so they stop surprising anyone.

Step 4: Build one calendar and feed it

Create a shared team calendar named something plain like “External dates.” Add each audited item as an event on its deadline, with two alerts: the start reminder from step three and a short-notice alert days before the deadline. Invite the owner and the deputy to each event so that one person on leave cannot hide a date. Put the first action and the link to the renewal location in the event description.

Where possible, export dates from the source systems rather than retyping them. Certificate inventories, cloud billing consoles and package manager lifecycle tools can all list expiries. Automation that posts upcoming expiries to a channel is a good complement, but keep the calendar as the human-readable single source.

Step 5: Add the planning dates

Beyond expiries, a handful of recurring external dates shape engineering work. Add them too.

  • Vendor-announced deprecation dates for APIs and SDKs you depend on.
  • Annual audit and certification windows that your organisation commits to.
  • Peak-traffic periods tied to the business, such as sales events or financial year-end processing, along with a change-freeze window around them.
  • Release dates of upstream projects you track, so that upgrade planning starts early.

For teams scoping a new piece of work, the thinking in an engineering-level view of web project requirements is a good prompt to ask which of those external dates a project inherits before it even starts.

Step 6: Test the reminders

Never assume an alert works. Create a dummy event for tomorrow, confirm it reaches both the owner and the deputy through the channels they actually read, and check that it survives a change of calendar settings. Wei’s team learned this the awkward way when a distribution list had been retired and half the reminders went nowhere.

Step 7: Review quarterly

Schedule a recurring fifteen-minute review. Scan the next ninety days, check owners are still employed and on the right team, and reconcile the calendar against the source inventories. Any entry without an owner is a bug; fix it that day.

A brief look outside engineering

Engineering is not the only field that lives by external dates. Finance runs on results days and filing deadlines, and the finance-education site Stocks Mantra lists a section of important dates for investors in its resources. It reads as plain education and says nothing about what anyone should buy or sell, yet it is a tidy example of the discipline we are describing: public dates, kept in one findable place.

Audit scorecard

After the first pass, a simple pass-or-fail review shows where gaps remain.

  • Every certificate has an owner, a deputy and a calendar entry.
  • Every licence has a renewal date and a cost note.
  • Every runtime and operating system in production has a recorded end-of-support date.
  • Every entry has two reminders, and a test alert has been delivered this quarter.
  • No entry is without a first action and a link.

What changed for Wei’s team

Within two quarters the team found eleven orphaned items, retired three, and handed the rest to named owners. The calendar did not make them heroes. It made them bored, which on-call engineers will tell you is the right outcome. A certificate renewal in the following year arrived as a Tuesday-morning task with a runbook link, and nobody opened a laptop at two in the morning.

If you take one thing from this checklist, take the order: discover, record, calculate, calendar, test and review. Start with certificates, since they cause the most dramatic failures, then widen the net to everything else with an end date. The first time a reminder fires and the answer is already written down, the effort repays itself.