Support and managed-service teams inherit systems they did not design, and they hear a lot of confident folklore about containers. Most of it is harmless, though a few stories quietly drive incidents, ticket volume and awkward weekend calls. The Ops Atlas includes a myths section in its entry on ContainerOps, its label for hands-on care of individual containers, images and registries, and it is worth reading before your next incident review. Below are six, each followed by the evidence and a replacement habit.
Consider the format a quick test: read the myth, ask whether anyone on your team believes it, and check the practical consequence in your own estate. The source is the Atlas entry that sets out ContainerOps and its myths.
Myth 1: Digest pinning shuts the door on security fixes
The belief: a digest locks the image in place, so patches stop arriving and pinning is irresponsible.
The evidence: according to the Atlas, that holds only when nothing is moving the pin. A digest identifies one exact build, which is what you want in production for predictability. Updates then arrive on purpose: a dependency bot proposes a pull request when a newer base appears, someone reviews it, and the new digest goes through testing and release.
What to do: pin everything, then automate the movement. Track base-image age to see whether the bot’s pull requests actually get merged.
Myth 2: Each container may only ever run one process
The belief: a container with two processes is wrong, full stop, and any exception must be refactored.
The evidence: the Atlas calls one process per container a good rule of thumb that is not a hard and fast rule. Splitting concerns usually helps with scaling and logging, but a small helper process or a log shipper in the same container is sometimes the pragmatic choice. What matters more is that something forwards signals and cleans up finished child processes, which is why the Atlas points to the init option.
What to do: treat the rule as a default, write down each deliberate exception and confirm that stops still work cleanly.
Myth 3: A clean scan report means the system is secure
The belief: if the scanner shows no critical findings, the container is safe.
The evidence: the Atlas says scanning alone is not security. A scanner only knows about weaknesses that have already been published, says nothing about how the container is run, and produces a list that is only useful when you can rebuild and redeploy quickly. A container running as an administrator with the privileged flag is not made safe by an empty report.
What to do: pair scans with fast rebuilds, non-root users, dropped capabilities and a measured time from fix to production. Scanner depth is covered by a sibling entry, so read this as the runtime half of the story.
Myth 4: A container is basically a tiny virtual machine
The belief: namespaces give each container its own machine, so an escape is unthinkable.
The evidence: containers share the host’s kernel. The Atlas calls that sharing a genuine exposure for untrusted code, which is the reason sandboxed runtimes exist. It also mentions two published runtime vulnerabilities, one from January 2024 involving leaked file descriptors and another published in November 2025, whose advice includes user namespaces that leave host root unmapped and rootless operation.
What to do: treat the container boundary as a useful fence, not a vault. Run unknown workloads in sandboxed runtimes and keep the runtime itself patched.
Myth 5: The container sees only the CPUs it was given
The belief: if a service is capped at two CPUs, the program inside will behave as if the machine has two.
The evidence: the Atlas observes that Python’s usual processor-count call still returns the host’s total from inside a container. Programs that size their thread pools from it launch too many workers and then spend their time paused. The Go language started sizing from the CPU limit only in version 1.25.
What to do: set worker counts by hand, then divide throttled periods by all periods to see the damage. For latency-sensitive services the Atlas aims for under 5 percent.
Myth 6: Our registry mirror is also our backup
The belief: since a pull-through cache keeps copies, losing the upstream catalogue is no problem.
The evidence: the Atlas states plainly that a cache is not a backup. A cache holds what has been pulled recently, can expire entries under policy and may not hold every tag you need. When a publisher moved free images off its original location in August 2025, teams that relied on caches discovered gaps.
What to do: keep your own copies of the images you cannot live without, push them to a registry you control and test a restore the way you would test any backup.
None of these myths is foolish. Each began as a reasonable simplification that was true in some setting, then travelled further than the setting. Correcting them is less about winning arguments and more about restoring the missing context.
A bonus myth: restarting fixes it, so nothing is wrong
Support teams know the temptation. A container misbehaves, a restart clears the symptom and the ticket closes. Occasionally that is the right call, but restarts also erase evidence: the in-memory state, the exact exit code in the moment and sometimes the only trace of a memory leak. Before restarting, capture the exit status, the last log lines and the limits in force. Five minutes of notes turn a recurring ghost into a diagnosable defect.
Why myths survive in support teams
Support work rewards speed, and folklore is quick. A confident sentence in a ticket can save ten minutes of reading, even when it is wrong. The antidote is a shared internal page that states each fact with its source and a date. If the page cites a reference and names an owner who updates it, people stop repeating half-remembered claims.
Our pieces on the principles behind effective delivery and the path from infrastructure as code to reliability engineering are good companions here, because each shows how habits outlast slogans. For a team that wants hands-on practice, a hands-on tutorial for practising DevOps at home lets people test each myth on a laptop in a quiet hour.
A ten-minute myth check for your next handover
When a new service lands on your desk, spend ten minutes on a short interview with its owner or its configuration. Ask what pins the image and what updates it. Ask which account the process uses inside the box and what would happen if that account were compromised. Look for ceilings on memory and process count. Watch one stop and see whether it ends politely or only after the forced kill. Finally, find where the logs go and whether anything ever deletes them.
Write the answers in the ticket. Each answer replaces a folk belief with a fact, and the next person on shift inherits something firmer than a rumour. After a few such handovers you will also have a small, honest picture of which myths your own estate actually suffers from, and that local picture beats any generic catalogue of folklore, this one included.