A Linux Host SysOps Audit Checklist You Can Run in an Hour

Rajesh Kumar

Rajesh Kumar is a leading expert in DevOps, SRE, DevSecOps, and MLOps, providing comprehensive services through his platform, www.rajeshkumar.xyz. With a proven track record in consulting, training, freelancing, and enterprise support, he empowers organizations to adopt modern operational practices and achieve scalable, secure, and efficient IT infrastructures. Rajesh is renowned for his ability to deliver tailored solutions and hands-on expertise across these critical domains.

An audit of a Linux host does not need a week or a specialised product. It needs a fixed list, an hour and a readiness to put every finding on paper. This checklist turns the practices of SysOps, the discipline of operating individual systems, into questions that can be answered with yes, no or a number. Run it on one representative machine first, then on the rest, and keep the results in a dated file so the next audit has something to compare against.

How to use the checklist

Work through each section in order. For every item record one of three outcomes: pass, fail or unknown. An unknown counts as a failure until someone can prove otherwise, since the aim here is to surface whatever has gone unexamined. The questions are drawn from the Ops Atlas entry on SysOps, and the section headings below match the areas that entry treats as in scope for a single host.

Section 1: Inventory and ownership

  • Is the host on a written list with someone accountable for it and a one-line reason it exists?
  • Is the operating system release still supported by its vendor? Record the end-of-life date on a calendar; the Atlas tracks the share of hosts past that date as a measure.
  • Can you reach the machine through a console or out-of-band path if the network configuration breaks?

Section 2: Access and privileges

  • Does every administrator use a personal account? A shared login with unlimited rights fails this item.
  • Do remote-access settings pass the daemon’s own syntax test, and does the printed effective configuration show the values you expect? Remember that for most keywords the first occurrence wins.
  • Are privilege-rule drop-in files named without dots, owned correctly and set to mode 0440? A file that is silently ignored provides no protection and no error.
  • If the host uses a cloud session tool, is a per-user “run as” setting configured, so sessions do not all land in one shared account?

Section 3: Scheduled work

  • Are recurring jobs defined as timers with catch-up for missed runs and a failure notification?
  • Where cron remains, are percent signs escaped, and is the shell assumption documented?
  • Can you see, for each job, whether it succeeded on its last run? The Atlas lists scheduled-job success as a headline measure.

Section 4: Logs and time

  • Is the journal’s storage mode set explicitly, and are its size limits chosen deliberately instead of inherited?
  • Do rotation settings avoid truncating files in place where a small loss of data would matter?
  • Is time synchronised, and have you decided how leap seconds are handled? The Atlas cites a configuration that smears the extra second over time and permits a stepped correction only in the first few updates after start.

The time item deserves emphasis. A network operator’s failure at the 2017 new year, caused by software assuming time cannot move backwards, affected a small fraction of DNS and web traffic and is documented in the Atlas as a case study. For readers who like to connect operational choices to organisational exposure, a practical discussion of reliability and risk in delivery provides useful framing.

Section 5: Packages, reboots and storage

  • Is there a recorded reboot policy, and how many hosts have updates waiting for a restart? The Atlas calls that backlog reboot debt.
  • Do you know how to detect a pending restart on this distribution, and can you roll back or pin a package if an update misbehaves?
  • Has the filesystem table been verified and a test mount run before the next planned reboot?
  • What is the disk runway? Take used space and growth per day, subtract the system’s reserved share and divide. Write the figure down. Under two weeks should open a ticket.

Section 6: Protection that stays on

  • Is the host firewall running, and are changes made with a scheduled automatic revert?
  • Is SELinux or AppArmor enforcing? Any exception should name one domain or one profile and carry a reason and an expiry date.
  • Has any baseline from a recognised benchmark been selected by the security function, and is the host’s conformance against it measured? Choosing the baseline is outside this audit; checking that the host follows it is inside.

Section 7: Recovery

  • Is there a backup, as opposed to a snapshot kept beside the data? The Atlas is explicit that these differ.
  • When was a restore last performed from that backup, and who signed the record?
  • Has a console drill been run recently? A large Windows outage in July 2024, which Microsoft estimated at 8.5 million devices, needed hands-on repair at the console, sometimes with disk-encryption recovery keys, and showed how much depends on having that path ready.

Section 8: Triage readiness

  • Is there a short, versioned triage script that captures load, memory, storage and network state before anyone restarts anything?
  • Are four basic alerts in place for the host, and is monitoring coverage recorded for every machine?
  • Do the on-call staff know how to read pressure figures, memory available and disk latency rather than relying on one headline number?

Scoring the result

Tally passes per section and list unknowns on their own. The Atlas ranks teams on a five-rung scale that begins at ad hoc and ends at replaceable, and a loose mapping from this audit looks like this: most fails in sections 1 to 3 mean the host is ad hoc, a clean 1 to 5 suggests baselined, and a documented recovery in sections 7 and 8 points toward measured or beyond. A perfect score is not the goal. Pick the three failures that would hurt most if they struck tomorrow, then give each a responsible person and a due date. Resist fixing cosmetic items first, since they raise the pass count without lowering the risk. Where a fix needs a change window, write the window into the ticket so the work does not stall waiting for a convenient moment. Keep the unknowns visible as well: each one is a small commitment to go and find out, and clearing them often reveals more than the failures do. A host with no unknowns is a host someone has truly looked at, which is already unusual.

If your organisation is thinking about its longer arc, an outlook on how DevOps may evolve in Japan and abroad offers context for where host-level discipline fits. For the underlying definitions and caveats, read the source entry on SysOps and adapt the questions to your own estate.

Repeat the audit each quarter. The first run finds problems; the second shows whether the fixes held; the third establishes a trend that you can present to stakeholders with confidence.

Leave a Reply