Get Free Quote

A Friday ContainerOps Checklist for Small Clinic IT Teams

Healthcare IT

Every clinic has someone who is not quite an IT department. Perhaps she is the office manager who set up the website years ago, or the part-time developer who also fixes the printer. This is the story of that person on a typical Friday, using a checklist built from the Ops Atlas, the reference glossary from xOps School. She is imaginary, and the setup is generic: a single server running the clinic’s public pages and a booking-request form, each in its own container. No part of this article touches clinical care.

The Atlas is clear that ContainerOps is not a standardised term and that unrelated projects reuse it. The Atlas gives it a working meaning: tending one container at a time, from traceable builds through careful naming and storage to well-behaved running. The ContainerOps entry lists what is in and out of scope, and the checklist below keeps to the “in” side.

Friday, 8:40: the coffee check

She starts the way she always does, with the status screen and a cup of tea. The first item on her list is simply to see whether anything restarted this week. A container that restarts quietly is telling you something, and the exit code is the clue. If she sees 137, the Atlas explains, that is a forced kill, 128 plus the kill signal 9. It can mean the stop request was ignored, or it can mean the kernel ran out of memory. Either way, she writes it down before she does anything else.

  1. Restarts and exit codes. Note any container that restarted, with the exit code and time.

Friday, 9:15: what is actually running

Next she compares the list of running images against her inventory page. Each row of the page holds an image, where it came from, when it was last rebuilt and who owns it. If something is running that is not on the page, she asks who started it. If something on the page is not running, she crosses it out with a date. Her rule: the page must match reality at least once a week.

  1. Inventory match. Running images equal the inventory, and each has an owner.
  2. Digest check. Every production container references a digest rather than a movable tag. The Atlas target for coverage is a hundred percent, and she keeps a running tally.

Friday, 10:00: how old is the foundation?

The Atlas names base-image age as a number worth watching. She looks at the build date of each base. If her developer’s bot has opened an update proposal that has been sitting for two weeks, she nudges him. She learned from the Atlas that pinning to a digest is only stale when nothing moves it, and that a bot proposing updates solves the problem.

  1. Base-image age. Anything older than the team’s agreed limit gets an update proposal reviewed.
  2. Runtime currency. Confirm the container runtime is still supported. The Atlas mentions that one common release line ended its support on 30 September 2026.

Friday, 10:45: resources and limits

Over a second cup, she looks at memory. The Atlas wants a memory cap and a process cap on all containers, so she checks that none slipped by. For CPU she is more cautious. The Atlas warns that a strict cap can throttle bursts of activity, and she watches the ratio of throttled to total periods, hoping to stay below five percent on the booking form because response time matters there.

  1. Limits present. Memory and process limits set on every container.
  2. Throttle ratio. Below five percent on latency-sensitive services.
  3. OOM kills. Count them against container-days; the Atlas lists the metric per thousand container-days.

Friday, 11:30: doors and keys

The next check is perhaps the most important. She lists published ports and asks of each: does this really need to be reachable from outside? According to the Atlas, published ports can bypass host firewall rules, and omitting an address binds every interface. She keeps administrative pages bound to the local address only. She also confirms that the Docker socket has not been mounted inside any container and that the list of people allowed to run Docker commands stays short. The Atlas equates both with root-level access.

  1. Ports. Only the reverse proxy faces the public network.
  2. Privileges. Containers run as ordinary users, privileged mode is off and no Docker socket is mounted.

Friday, 13:00: housekeeping after lunch

The afternoon starts with disk space. Two things fill disks quietly: logs and old images. Left at its default, the logging mode in use never rotates its files, as the Atlas notes, so she makes sure a size limit exists. On images, she confirms the registry has a retention rule and that garbage collection has run recently.

  1. Logs. Rotation set, and disk usage under the team’s threshold.
  2. Registry tidy. Retention policy applied and old versions removed.

Friday, 13:45: the supplier question

Once a month she adds a thirteenth thought that does not fit on the list: where do our images come from, and what would we do if one source changed its rules? The Atlas describes an example from 2025, when a popular maker of ready-built images moved its free catalogue off its original public home, stopped updating the old one and told users to revise pipelines and mirrors. She has since added a column to her inventory page that shows the second source for each important image. It took twenty minutes and removed a whole category of worry.

She also checks the pull allowance. The Atlas puts the anonymous allowance on a major public registry at a hundred pulls in six hours, so she keeps the clinic’s builds signed in. It is a tiny detail that stops a deployment day from turning into a mystery.

Friday, 14:00: a minute for the people side

Checklists work only if they are short, and hers is twelve lines on a single page. She keeps a small log of results, one line per Friday. Over time the lines turn into a trend that she can show anyone who asks how the system is doing. She also notes any item she skipped and why, because skipped items are the first to become habits.

The checklist grew out of her reading about a thirty, sixty and ninety day approach in the Atlas. In the first month she took the inventory and set log defaults. In the second she moved to digests and bot updates. In the third she made the checks blocking and rehearsed a rebuild. She is proud of the rehearsal, because the first attempt revealed that nobody had written down how to restore the booking form’s configuration. Writing it down took an hour.

Friday, 15:30: what if something is off?

When an item fails, she does not patch a running container by hand. Editing a live container by hand is something the Atlas warns against, because a restart wipes the repair and leaves no record. She rebuilds the image with the fix, tests it, deploys by digest and records the change. If the new build misbehaves, the previous digest is one line away.

She also keeps two links in her notes for the rest of the team. One is a general guide on how to choose eye hospitals, which she uses as a reminder that readers of the clinic’s site should find honest, plain information. The other is a practical routine for building a pocket money plan with an expense tracker, which she shares with new staff who want a simple model for keeping records every week.

Friday, 16:30: sign-off

Twelve checks, about ninety minutes, one line in the log. She closes the status screen, locks the cupboard where the spare router lives, and leaves. The best compliment her setup ever gets is silence: no emergency calls over the weekend, no mystery restarts, no surprise email from the registry about limits.

If you are the person in your clinic who inherited the website, copy the list and change what you need. The point is not to match anyone’s list exactly, but to have one, to run it at the same time each week and to write down what you find.