Four Ways to Deliver Patches Compared: Strengths, Gaps and Fit

Four Ways to Deliver Patches Compared: Strengths, Gaps and Fit

When a team asks how it should deliver security and bug-fix updates, the answers tend to arrive as brand names. A more useful starting point is the delivery pattern behind the tool. This comparison sets four common patterns side by side: unattended in-place updates, a ringed managed rollout, live patching, and rebuild-and-replace. For each one it asks the same practical questions: what does it do well, where does it fail, and who should choose it? The framing leans on what the Ops Atlas says about PatchOps and on the NIST guidance behind it.

The yardstick: what a good delivery pattern must provide

Before comparing, fix the criteria. The Atlas treats patching as more than installing files. A pattern should handle cadence, tell you which machines an update applies to, support staged rollout, deal with restarts, allow some form of recovery when things go wrong, and produce proof. We will score each approach against six simple questions.

  1. Can it stage changes so a problem hits few machines first?
  2. How does it deal with restarts?
  3. What recovery path does it offer?
  4. Does it create evidence you can show someone else?
  5. How much upkeep does it need?
  6. Which machines suit it?

Option one: unattended in-place updates

This is the simplest pattern. A helper on each machine fetches and installs updates on a schedule, with no human involved. Two familiar Linux examples, both named in the Atlas, are unattended-upgrades for the Debian family and dnf-automatic for the Red Hat family.

Strengths: near-zero setup, no central server to run, and it works on a lone machine in a cupboard. For a small estate of non-critical hosts it is often enough.

Weaknesses: every machine updates on its own clock, so there is no first ring and no pause between waves. Restarts follow local settings and may surprise people. Evidence is whatever you scrape from each host afterwards. And if a bad update lands, it lands everywhere at roughly the same moment.

Choose it when the machines are disposable, similar and low in consequence, and when a person will still count the results.

Option two: a managed, ringed rollout

Here a central service decides which machines get which updates and when. The Atlas lists cloud-provider managers for patching virtual machines, a desktop-management suite that has built-in ring defaults, and an older server update service whose deprecation was announced in 2024 although it remains supported. The defining feature is the ring structure: a small canary group, a pilot group, a broad wave and a careful group for sensitive systems.

Strengths: staged rollout with gates, deadlines, compliance reports and a single place to see progress. Emergency updates can use a fast lane with a shortened canary period.

Weaknesses: it needs design. Groups must be defined, windows negotiated, ring captains named and gates written. The central service is a sensitive target in its own right: whoever controls it controls admin-level execution across the whole fleet. And dashboards can flatter you: coverage counted only among reporting machines hides the silent ones.

Choose it when you run dozens or thousands of machines, answer to auditors, or have customer-facing systems where a bad update hurts.

If your team already uses automation to cut incident response time, this choice will feel familiar. Our walk-through of speeding up incident handling with operations analytics shows how routine work and exceptional work can share the same data.

Option three: live patching

Live patching slips chosen fixes into software that is still running, sparing the reboot. The Atlas points to one Linux vendor’s livepatch service and to a hotpatch option for Windows Server hosts enrolled via Azure Arc, which cost nothing extra in October 2026.

Strengths: fewer disruptive reboots, shorter exposure windows and happier application owners. It suits machines with strict uptime needs.

Weaknesses: coverage is partial, since not every fix can be applied live, and the system still needs periodic restarts for the rest. It complements the other patterns instead of replacing them. It also adds one more thing to monitor.

Choose it when restarts are expensive and you can combine it with a managed rollout for everything it does not cover.

Option four: rebuild and replace

In container and image-based setups, you do not alter a running machine. You build a new image containing the fixes, test it, and swap it in. The Atlas is firm on one point here: the claim that containers mean no patching is a myth, because rebuilding is patching. It still wants a calendar, a due date and evidence that stale images have disappeared.

Strengths: consistent results, easy rollback to the previous image, and a natural canary: send a little traffic to the new version first.

Weaknesses: image rebuilds are a separate discipline with their own tooling, and the Atlas places them outside the core scope of PatchOps. Old images can linger in registries and run for months. Third-party base images add a dependency on someone else’s schedule.

Choose it when the workload is stateless or easily replaced and your pipeline already produces images routinely.

Hidden costs that rarely appear in comparisons

Every pattern has a cost that sits outside its feature list. Unattended helpers cost you visibility: the day something breaks, you will wish you had central logs. Managed rollouts cost design time and the discipline to keep the group definitions current as machines come and go. Live patching costs attention, because you now track two kinds of fix state per host. Rebuild pipelines cost engineering effort and storage, and they demand someone to retire old images deliberately.

A fair question to ask of any pattern is what happens when the person who set it up leaves. Patterns that rely on written gates, named captains and a visible register survive that moment. Patterns that live in one engineer’s scripts do not, however elegant they look on the day.

Side-by-side summary

  • Staging: in-place helpers have none; managed rollouts have rings; live patching inherits whatever the host uses; rebuilds stage through traffic shifting.
  • Restarts: in-place follows local settings; managed rollouts schedule them; live patching reduces them; rebuilds replace instances.
  • Recovery: in-place has no clean undo for security fixes; managed rollouts mostly roll forward with a corrected update; live patching can be disabled for some fixes; rebuilds revert to a prior image.
  • Evidence: in-place is scattered; managed is centralised; live patching adds status per host; rebuilds produce image records.
  • Effort: in-place is lowest; managed rollouts and rebuilds take the most design.

Mixing patterns sensibly

Real estates rarely fit one pattern. A typical mixed design runs unattended helpers on a few disposable hosts, a managed rollout across the main server estate, live patching on the machines that cannot reboot easily, and rebuilds for containerised services. The danger is that each pattern reports in its own format. Choose a single coverage definition for all of them, with every in-scope asset in the denominator and credit given only to hosts that have restarted where needed and still report in.

For teams building a longer-term operations plan, our overview of a practical operations roadmap for infrastructure teams offers a way to sequence these decisions without trying to do everything in one quarter.

A quick decision guide

  1. Fewer than twenty low-consequence machines: start with unattended helpers and a monthly manual count.
  2. A growing mixed estate: adopt a managed rollout and define rings.
  3. Uptime-sensitive servers: add live patching to the managed rollout.
  4. Container platform: rebuild on a cadence and verify old images disappear.

Whatever you choose, keep proof in the loop. The full set of definitions, standards and examples is in the Atlas page on PatchOps, and it rewards a careful read before you commit to tooling. Patterns matter more than product names, and the sooner a team can describe its pattern in one sentence, the sooner it can improve it.