Back to Articles
openvox November 18, 2026

The Work You Can Still Do in a Change Freeze

JS
Jerald Sheets

Catalog Cadence, issue 2. First published on LinkedIn, 18 November 2026: read it there and subscribe.

If your change calendar isn't frozen yet, it probably will be by next week. That's the correct call. Thanksgiving week is a terrible time to cut over a compiler, bump a CA, or "quickly" swap a package on the server that signs certificates. The on-call person would like to eat dinner.

A freeze is not a week off for the estate. The work that is safe this week is the work that does not require a production change window: looking, counting, writing things down, and tightening the feedback loop on a laptop. That is also most of what a good Healthcheck is made of.

Here is the freeze-week list I actually use.

Inventory the estate you have, not the one you remember

You can't plan January for a fleet you can't describe. Before anyone is fully out, I want a written snapshot:

  • Server, agent, and PuppetDB or OpenVoxDB versions, plus the oldest agent still checking in.
  • The OS spread, including the oldest OS still managed. OpenVox 9, released in October, no longer builds packages for Enterprise Linux 7, Amazon Linux 2, Debian 11, Fedora 42, or Ubuntu 25.04. Those agents stay on OpenVox 8 for now, or the OS gets fixed first.
  • Where server and ca_server point on the agents. A VIP? The CA host directly? A DNS alias somebody set up years ago that nobody remembers?
  • How code gets deployed: r10k, Code Manager if you're on PE, or "git pull on the server." No judgment. Okay, a little judgment.
  • How nodes get classified: Hiera plus roles and profiles, an ENC, PE Console groups, or regex in site.pp.
If you have PuppetDB or OpenVoxDB, a query against the inventory endpoint will give you certnames and facts.puppetversion in one pass. Save the output. January-you will not remember what November-you "kind of knew."

Back up the things that are painful to rebuild

A freeze is an excellent time to confirm that the backups you think you have are actually restorable.

Back up all of /etc/puppetlabs/ on the CA, every compiler, and the database host, and take a PuppetDB or OpenVoxDB dump. Yes, even if "backups run every night." Confirm the last successful dump is recent, is not zero bytes, and can be read by someone other than the person who set it up. If the CA were lost on Friday afternoon, who would notice, and what would they open first?

Write that down where the on-call person can find it without a phone call. A one-page "if the CA is unhappy" note has ended more holiday incidents than any clever module I have ever written.

Let the robots do the nagging

Style nits and parse errors do not need a change window. They need to die on a laptop.

Run openvox-lint (or puppet-lint) and puppet parser validate across the control repo and in-house modules. Don't fix every arrow-alignment warning on Wednesday afternoon unless you enjoy that sort of thing. Do count the errors, note the modules that are on fire, and — if you haven't already — put both checks in a git pre-commit hook and in CI.

Misaligned arrows, quoted booleans, and trailing whitespace should get caught before they ever see a pull request, let alone a 2 a.m. deploy. Catalog compile failures for your top roles should get caught in CI, against the version you actually run. "It parses on my workstation" is not a pipeline.

A crude first pass for legacy facts and old function calls is also fair game this week:


grep -rnE '\$::?(operatingsystem|osfamily|hostname|fqdn|ipaddress|lsbdistcodename)\b|hiera\(' site/ modules/

It isn't gospel. It is an honest first pass. The hits you find are work you can ticket for after the freeze, not work you should merge on Wednesday night.

Write down how a change reaches a node

If I could ask only one deliverable of a quiet week, it would be this: a one-pager that says how a change gets from a commit to a catalog on a node.

What branch is production? Who can merge? What CI must go green? How does r10k or Code Manager deploy an environment? How would you roll back? Do you have a --noop habit before a real run on a new profile? If the person who usually does this is on a flight to see family, can someone else follow the page without a phone call?

While you're at it, write the classification story in one paragraph. If you can't, that's the finding. It is also one of the first things we look at in a Healthcheck.

What not to do this week

  • Don't change vendor and major version in the same window, and definitely don't do it this window. A Puppet-to-OpenVox move or an OpenVox 8-to-9 jump can wait until the calendar is boring again.
  • Don't sign a pile of outstanding certificate requests "to clear the queue" unless you know why each host requested one.
  • Don't run a failover test the night before Thanksgiving. That's a short story with a bad ending.
  • Don't confuse a clean lint run with a healthy estate. Lint is necessary. It is not the architecture review.

If you want a second set of eyes while the calendar is quiet

A Healthcheck is designed for exactly this kind of week: we read the code, we don't touch production, and we hand you a written report of what's healthy, what's risky, and what we'd redesign first. It is a fixed price of $1,495. Onsite is available if you'd rather do it in a room with a whiteboard after the holiday — ask us and we'll work out the details.

The report is yours. Some teams take the to-do list and run with it in December and January. Others come back to us for on-prem repair or a migration sprint, or for Configuration Management as a Service when they don't want the estate living in one person's head through another holiday season. Those are later conversations. The starting point is the look.

If a quiet week is a good week to start that look, book a 15-minute intro call or visit ssconsultinggroup.net. We reply within 24 hours.

Next issue lands the day before Thanksgiving, so I will not suggest you start a migration. I will tell you why we treat OpenVox as the true open-source path, when staying on Puppet is the honest call, and how to move without a mass re-signing — after the freeze, not during it.

— Jerald Sheets, S & S Consulting Group

---

Related Articles

Need Help With Your Infrastructure?

We can help you implement the patterns and practices discussed in our articles.

Get in Touch