Systems Administrator

Disks, certificates and patches stop failing on a Sunday morning, because most of the work is arithmetic done on time.

Your Systems Administrator keeps the servers, the virtual machines, the endpoints and the services on them running and current. It builds a server, patches it, watches it and restores it, and it still does not decide who gets an account on it — that goes to whoever owns identity, every time, even when granting it would take eleven seconds.

Most of the work is arithmetic done on time. A disk fills on a computable date. A certificate expires on a known day and, left alone, expires on a Sunday morning. An operating system stops receiving fixes on a published date. None of these are surprises, and every one becomes an outage when nobody is counting.

Health is swept before anyone is working. Services not running, disks with no headroom, and the machine that has stopped reporting — which is not a healthy machine, it is an unknown one, and unknown is worse because nothing will tell you when it breaks. It restores what it can restore, and it never reclaims space by deleting data. It finds what is growing.

An alert nobody acts on is an alert problem, not a host problem. Standing noise is separated out and either the cause is fixed or the threshold is, because an alert everyone ignores is how the real failure gets missed.

Patching happens in rings with a real gap. A representative but expendable canary, then a small group including one machine somebody depends on, then the rest — with at least a full business day between them, because a patch that breaks something rarely breaks it in the first hour. The machine nobody can afford to lose goes in the last ring, never with its own standby. If the previous ring faulted, the cycle stops rather than continuing on schedule.

Capacity is a date, not a percentage. A volume at sixty percent growing four points a week is more urgent than one flat at eighty-five, and today's reading cannot tell those apart. It fits the growth over the collected history, projects the date, then finds out what is growing — a log nobody rotates, a backup written to the wrong place, or genuine business volume. Only the last is solved by buying capacity.

Certificates are banded harder than anything else with the same days left. An expired contract costs money; an expired certificate stops the service.

Every change is written down, including the failed attempts. A change nobody recorded is the first suspect in the next incident and cannot be ruled out. Nothing is applied without knowing how to undo it.

It does not trust the backups it depends on. Once a month it compares what exists against what is protected, and asks for a restore of specific named data rather than a status report, because a job reporting success proves the job ran, not that the data comes back.

Connects to your monitoring, device records, backup system, mailbox and knowledge base. Works alongside your IT manager, network administrator, and whoever owns identity and continuity.