GoldenrozBook a call

Server maintenance checklist for a server room

Server maintenance is a schedule of weekly, monthly, quarterly and yearly tasks with a named owner for each. Check backups, alerts and disk health every week, apply operating system patches every month, and test a full restore, update firmware and inspect the UPS and cooling every quarter. Once a year, test disaster recovery, review hardware warranties and operating system support dates, and update the documentation.

A gold sphere under a clear glass dome on a navy base.

How to use this checklist

The lists below are grouped by how often a task should happen. They assume a typical on-premises environment: physical hosts running a hypervisor, Windows and Linux virtual machines, shared or local storage, a backup system, and a server room with a UPS and cooling.

Two rules make any checklist work. Every task has a named owner, and every completed task leaves a record. A checklist nobody signs is a wish.

Weekly

  • Backups. Review every job, including the ones that report success. Look for jobs that ran much shorter or much smaller than usual.
  • Alerts. Clear or act on every monitoring alert. Alerts that are ignored each week should be fixed or removed, because noise hides real problems.
  • Disk space. Check volumes above your warning threshold and find out what is growing.
  • Hardware status. Review the out-of-band management controller on each host for failed or predicted-fail disks, power supplies, fans and memory errors.
  • Storage and RAID. Confirm arrays are healthy and no rebuild is running unnoticed.
  • Security events. Review failed administrator logins, new administrator accounts and antivirus or endpoint detections on servers.
  • Server room walk-through. Look at indicator lights, listen for failing fans, read the temperature display, and check the UPS panel.

Monthly

  • Operating system patches. Apply to a pilot group, wait, then apply to production in a maintenance window. Reboot. A patch that is installed but waiting for a reboot is not applied.
  • Hypervisor and management tools. Check for security updates.
  • Application updates. Database engines, web servers, backup software and agents.
  • File restore test. Restore a file or a mailbox item from backup to prove the chain works.
  • Capacity review. CPU, memory and storage trends per host and per cluster. Note anything that will run out within a year.
  • Account review. Disable accounts of people who left. Review who is in administrator groups.
  • Certificate expiry. List certificates that expire in the next 90 days.
  • Time synchronization. Confirm servers agree on the time. Authentication and logs depend on it.
  • Snapshots. Delete old virtual machine snapshots. They are not backups and they grow.

Quarterly

  • Full restore test. Restore a whole server to an isolated network, boot it, and confirm the application starts. Record how long it took.
  • Firmware. BIOS or UEFI, management controller, RAID controller, network cards, disks, storage arrays and switches. Read the release notes and follow the vendor’s supported order.
  • UPS. Run the self-test, read the battery health and runtime estimate at current load, and check the event log.
  • Cooling. Clean or replace filters, check condensate drains, confirm the temperature alert reaches a person.
  • Failover test. Move workloads between cluster hosts. Pull one power cord on a dual-supply server during a window to prove both feeds work.
  • Firewall and remote access review. Remove rules and accounts nobody can explain.
  • Vulnerability scan. Scan servers and fix or document the findings.
  • Physical cleaning. Dust in air intakes raises temperatures and shortens component life.

Yearly

  • Disaster recovery test. Recover the systems on your priority list at the recovery site or in the cloud, with the people who would do it in a real event.
  • Lifecycle review. For every server: warranty end date, operating system support end date, hypervisor support status.
  • UPS batteries. Compare age with the manufacturer’s replacement guidance. Schedule replacement before failure.
  • Generator. If you have one, test under real load. An idle start proves little.
  • Documentation. Update the server inventory, network diagram, rack layout, backup scope and recovery runbooks.
  • Credentials. Rotate service account and break-glass passwords. Confirm the emergency credentials are where the runbook says.
  • Licenses and support contracts. Renew what you use and cancel what you do not.
  • Access list. Review who holds keys or badge access to the server room.

Firmware and patching: order and rollback

Patching causes outages when it is done without a way back. Before each window:

  1. Confirm last night’s backup of the systems being patched completed.
  2. For virtual machines, take a snapshot before the change and delete it once the change is verified.
  3. Patch one node of a cluster, verify, then patch the next.
  4. Have the management controller reachable, so a server that does not come back can be seen and fixed remotely.
  5. Write down what “working” means for each application, and test it after the reboot.

Firmware deserves more care than operating system patches. Vendors publish supported combinations of firmware, drivers and hypervisor versions. Follow the matrix. Jumping several versions at once on a storage controller is the kind of change that deserves a full backup and a longer window.

Hardware lifecycle and Windows Server end of life

A server has two clocks. One is the hardware: warranty, spare parts and firmware updates. The other is the operating system. When either runs out, the server is a risk even if it runs perfectly.

Microsoft’s Fixed Lifecycle Policy gives Windows Server versions a mainstream support period followed by an extended support period. Security updates continue through extended support and stop at its end. The dates below are from Microsoft’s product lifecycle pages on learn.microsoft.com, as shown in November 2026.

Version Status in November 2026 End of extended support
Windows Server 2012 R2 Out of support. The final year of paid Extended Security Updates ended October 13, 2026 October 10, 2023
Windows Server 2016 Extended support January 12, 2027
Windows Server 2019 Extended support January 9, 2029
Windows Server 2022 Extended support. Mainstream support ended October 13, 2026 October 14, 2031
Windows Server 2025 Mainstream support, listed to run into late 2029 Listed in late 2034

Two of these matter now. Windows Server 2016 has about two months of support left at the time of writing, and Microsoft’s lifecycle page lists paid Extended Security Updates for three further years. Windows Server 2012 R2 has no remaining update path. Always confirm dates on the lifecycle page for your exact product and edition.

Applications follow their own clocks. Microsoft lists the end of extended support for Exchange Server 2016 and 2019 as October 14, 2025. If mail is the reason an old server is still running, the Office 365 migration checklist covers the way off it.

An operating system upgrade is rarely only an upgrade. The application vendor must support the new version, and the hardware may not. That is why the lifecycle review belongs in the yearly list: it produces next year’s project list and budget. Planning those replacements is part of IT infrastructure services.

The room itself: power, cooling and access

Server room design problems show up as maintenance problems, so the room is on the checklist.

Power. The UPS must carry the full load with headroom, and its runtime must cover either a generator start or an automated clean shutdown. Test the shutdown automation. Dual power supplies should be connected to separate feeds, or they provide no redundancy.

Cooling. Use dedicated cooling that runs independently of the office air conditioning schedule. Keep intake temperature inside the range your hardware vendor publishes. Arrange racks so cold air enters the front and hot air leaves the back without mixing. One cooling unit is a single point of failure, so decide what happens when it stops.

Monitoring. Temperature, humidity, water leak and door sensors should send alerts to people, including at night.

Access and order. Lock the room and log entry. Label both ends of every cable. Keep cardboard, spare furniture and cleaning supplies out. A room that doubles as storage gets bumped cables and blocked airflow.

Documentation: the cheapest maintenance task

When a server fails at night, the time to recovery depends on what is written down. Keep, at minimum:

  • An inventory of every server: role, owner, operating system, warranty end, support end
  • Network and rack diagrams that match reality
  • Backup scope, retention and the restore procedure
  • A recovery priority list agreed with the business
  • Vendor support contacts and contract numbers
  • A change log of what was patched or changed, and when

Store a copy somewhere that does not depend on the servers it describes.

Backup is not disaster recovery

The weekly and quarterly backup checks prove that data can be restored. Disaster recovery answers a larger question: where systems run if the room is unavailable, in what order they come back, and how long the business can wait. Those targets drive the design and the cost, and they are covered under disaster recovery services.

What to do next

Take the yearly lifecycle review first, because it has deadlines. List every server with its operating system version and warranty date, and mark anything on Windows Server 2016 or older. Then assign owners to the weekly and monthly lists. If you want the schedule run for you, our server management services cover patching, firmware, monitoring, backup checks and documentation as part of a managed IT services retainer, quoted after an assessment. Replacement and migration projects start at $25,000 at a fixed price per phase.

Questions buyers ask

How often should servers be patched?

Monthly is the common baseline for operating system updates, because Microsoft releases security updates on a monthly cycle. Patch a test or low-risk group first, then production a few days later. Vulnerabilities that are being actively exploited should be patched outside the normal cycle.

When does Windows Server 2016 reach end of life?

Microsoft's lifecycle page lists the end of extended support for Windows Server 2016 as January 12, 2027. After that date it receives no security updates unless you buy Extended Security Updates, which the same page lists for three further years. Confirm the date on Microsoft's lifecycle page before planning.

How long should a physical server stay in service?

There is no fixed number. The practical limits are the manufacturer's warranty and support contract, the availability of spare parts and firmware updates, and whether the hardware can run a supported operating system and hypervisor. Plan replacement before the support contract ends, not after the first failure.

What should a server room have?

Controlled access, dedicated cooling sized for the heat load, a UPS sized for the equipment with enough runtime for a clean shutdown or generator start, environmental monitoring that sends alerts, labeled cabling and no water pipes or storage boxes. Fire detection and suppression follow local codes.

Is a backup job that reports success enough?

No. A successful job means files were written, not that a system can be restored from them. Run a real restore on a schedule: a file each month, a full server each quarter, and time how long it takes. A backup that has never been restored is unproven.

Written by

CTO and technology partner

Shai leads the technology side: business software, SaaS and mobile apps, payments and escrow systems, Postgres security, AI agents built with Claude Code and MCP, and IT and infrastructure projects with senior infrastructure engineers.

Keep reading

Talk to the person who will do the work

One hour, no sales team. You talk to Daniel or Shai, we look at your account or your product, and you leave with a clear next step.

Or send the details

Book a 1-hour consultationWhatsApp us