If server room cooling fails, protect people first, then limit IT damage by understanding how fast temperatures are rising and which loads can be shed safely. Check obvious faults such as tripped power, blocked airflow, full condensate pumps and closed valves, escalate to competent engineers, and deploy temporary cooling only where electrics and drainage allow. A written playbook tested in advance is far more effective than improvised decisions during an incident.
Prioritise safety and situational awareness
Do not enter plant areas if there is suspected refrigerant leakage, electrical burning, flooding or another immediate hazard. Raise the alarm through the organisation's normal emergency route and keep untrained staff clear of live equipment and water near electrics.
From a safe position, note inlet temperatures, rate of rise, which cooling units are offline, and whether UPS or generators are also abnormal. Time to thermal limit can be short in dense rooms, so early facts matter more than perfect diagnostics.
- Confirm who is in charge of facilities and who owns IT shutdown decisions.
- Record temperature trend, not only a single reading.
- Check whether alarms reached on-call staff as designed.
- Avoid opening doors in a way that spreads smoke, water or heat unpredictably.
Run a rapid technical triage
Look for recoverable faults: isolators off after maintenance, tripped breakers, dirty filters starved of airflow, outdoor units blocked, condensate pumps in fault, or controls left in an unexpected mode. Only competent persons should reset or open refrigerant and electrical systems.
If one unit of an N+1 pair has failed, confirm the surviving unit is actually carrying the load and that hot spots are not forming. Do not assume redundancy is working without inlet measurements.
Reduce heat and buy recovery time
IT teams should follow a pre-agreed shedding order: non-production hosts, batch jobs, development kits and other deferrable loads before business-critical services. Random unplugging can create worse outages than a controlled reduction.
Temporary cooling may help if power, exhaust or heat-rejection paths and condensate management are ready. Portable units without a planned exhaust route can worsen conditions. Keep fire routes clear while temporary equipment is in use.
Recover carefully and learn from the incident
After plant is restored, watch inlet temperatures back into the agreed envelope before restoring shed loads. Investigate root cause: capacity shortfall, airflow blockage, maintenance error, power loss or control failure.
Update the playbook, alarm contacts and maintenance regime. If the building's air conditioning systems fall under the statutory inspection regime above 12 kW effective rated output, keep that compliance activity separate from emergency repair, but ensure records of the installed systems remain accurate after changes.
Frequently asked questions
How fast can a small server room overheat?
It depends on IT load, room volume and residual airflow. Dense cupboards can climb quickly, which is why rate-of-rise monitoring and rehearsed responses matter.
Should we open the server room door immediately?
Only as a judged temporary measure. It may dump heat into occupied areas and provide little relief if the corridor is already warm. Follow the incident plan rather than habit.
Who decides to shut servers down?
That should be defined in advance between IT and facilities. The cooling incident plan needs named roles, not improvised authority during a crisis.
What should be reviewed after the event?
Root cause, alarm performance, spare capacity, airflow condition, temporary-cooling effectiveness and any changes needed to maintenance or redundancy.
Related guides
Sources and further reading
How this guide was prepared
This guide was written by the Local AC Installers Editorial Team. We compare official guidance, legislation and established consumer information, separate general information from project-specific advice, and show publication dates and sources. We do not accept installer accreditation claims at face value.