Get Free Quote

Six SysOps Myths Weighed Against the Documented Record

Healthcare IT

Operations folklore travels faster than operations evidence. Anyone who has sat in a late-night call about a stuck booking system or a missing report has heard confident advice that turns out to be wrong. Below, six widespread beliefs about server care, the field called SysOps, are checked against what the Ops Atlas documents. The setting is a medical-office back end, but nothing here is clinical: it concerns only the computers that carry scheduling, documents and web pages.

For each myth we state the claim, say why it is tempting, and give what the documented evidence says. Where the Atlas supplies a number or a date, we quote it exactly and leave it unadjusted.

Myth 1: “When in doubt, restart it”

Why it is tempting: restarts often work, and they are quick. A restart clears stuck states, and the pressure to restore service is real.

What the record says: the Atlas names restart-first behaviour as an anti-pattern because it destroys evidence. What was in memory, which connections were open and the order of recent events all disappear just when you need them. The recommended order is to capture first: note what is busy, what is waiting and what the logs say, then decide. A restart remains a valid tool; it should be a deliberate step, taken after the evidence is saved, and recorded with the time, the reason and the outcome so that a repeat visit to the same fault can be compared with the first. Note the cost of getting this wrong: if the fault returns next week, you have nothing to compare.

Myth 2: “Turn off the security layer to get past the error”

Why it is tempting: mandatory access control and host firewalls generate cryptic refusals, and switching them off makes the symptom disappear.

What the record says: the Atlas flags the habit of dropping the firewall, SELinux or a sandbox so that something will run. It points to narrower alternatives. For SELinux, a single domain can be set to permissive with one command; for AppArmor, a single profile can be put into complain mode. In both cases the rest of the host stays enforced, and the logs record what would have been denied, which is the information you need to write a proper rule. The Atlas advises against relaxing either layer for the whole host.

Myth 3: “Protect the important process so the system never kills it”

Why it is tempting: when memory runs out, the kernel picks a process to terminate, and nobody wants it to pick the database.

What the record says: setting the extreme score adjustment that exempts a process, the value of minus 1000, does not remove the pressure; it moves the kill elsewhere. Something else is terminated instead, and it may be as valuable as the process you shielded. The Atlas files it under anti-patterns and points instead to why memory ran short in the first place. Related caution: a small “free” figure is not the problem by itself, because the page cache fills spare memory, and the number to read is the available figure.

Myth 4: “The disk shows 100 per cent busy, so it is saturated”

Why it is tempting: the busy-percentage column in common tools is easy to read and looks like a fuel gauge.

What the record says: the Atlas warns against trusting that figure on SSDs and RAID sets. Such hardware serves requests side by side, which means a drive can report itself fully busy while capacity remains. Look instead at how long requests wait, how deep the queue is and whether errors are reported. The Atlas supports this with its broader triage method: for each resource, ask how busy it is, how much work is queued and whether it reports errors.

Myth 5: “Log rotation is harmless housekeeping”

Why it is tempting: rotation looks like a tidy-up with no downside, and the default settings are rarely questioned.

What the record says: rotating by copying and then truncating can drop a sliver of log lines written in the gap, which the Atlas describes as a very small time slice. For routine access logs that cost is trivial; for records you must preserve, choose a method that signals the program to reopen its file. The Atlas adds a point about the system journal: its defaults let it take about ten per cent of the filesystem, keep fifteen per cent free and limit each file to 4 GiB, so set the storage behaviour explicitly rather than leave it implicit.

Myth 6: “Every platform can undo an update”

Why it is tempting: if your package manager at the office has an undo command, you assume all of them do.

What the record says: the Red Hat family’s package tool keeps a transaction history that can be reversed. Debian-style apt has no such undo, and the Atlas’s workaround is to install and pin an older version. A team that mixes platforms should write down which behaviour applies where, and take a snapshot or backup before any update that matters. Related: confirm whether the machine needs a restart afterwards, since a command on Red Hat-type systems signals it by exit status 1, and Ubuntu creates a marker file.

The cost of getting it wrong

None of these myths is silly; each solves a real irritation. The damage shows up later, when a team that has rehearsed shortcuts has to explain an outage with no logs, no baseline and no record of the last change. A useful exercise is to take one recent incident and ask which of the six shortcuts was used, and what it removed from view. In most cases the answer points to a small procedural fix, such as a capture step before any restart or a note field in the change log, rather than to new software.

What the six share

Every one of them buys quick relief at the price of information: a restart erases clues, a disabled safeguard hides the real rule, an exempted process shifts the damage, a misread gauge sends you after the wrong resource, a careless rotation drops data and an assumed undo fails when needed. The remedy in the Atlas is a habit rather than a product: note what you observe, alter a single item per step, and log each alteration.

A quick way to apply this

Before the next incident, prepare a one-page triage sheet: the commands that show load, memory availability, disk state and failed units, plus the place to save the output. Practise it once, on a quiet afternoon, with a colleague timing you; the first run will expose missing commands, unclear ownership and permissions you did not know you lacked. The routine resembles the checklists in a piece on preparing a document checklist before a bank visit, where having the list ready is half the work. A parallel in planning routines appears in a short guide to keeping a month-long expense routine going: small, repeatable steps survive when heroics do not.

For the complete reference, including scope, metrics and the cases behind these myths, read the Ops Atlas page on SysOps. Treat it as a source to test your own habits against, and update your assumptions when the evidence says so.