Summer in Germany – that usually means more than outdoor pools and ice cream; it also puts our hardware to a real test. If you run your own servers or a homelab, you're probably glancing at your monitoring dashboards a little more nervously these days.
I saved the readings from my server "HorstTower" for the week from 21 to 28 June. The result? The heatwave that had us sweating outside that week shows up beautifully in the system data – including a classic "layer 8 error" on my part in the middle of the weekend and a proper rush on Sunday.
CPU load: a perfectly normal (and an extremely busy) day
Let's look at the raw compute first. Over the week, the CPU load shows the familiar picture of a system in dynamic use:
During the week (21 to 25 June) we see a solid base load of around 10 to 15 per cent, interrupted by regular short spikes. A few striking outliers stick out, but overall the system runs stably.
Things get really exciting on Sunday, 28 June: the place is on fire – quite literally. For long stretches the system is under heavy load, with utilisation constantly jumping up close to the 100 per cent mark. But did you notice Saturday? More on that in a moment.
CPU temperature: when the baseline creeps up
It gets interesting when we compare that load with how the CPU temperature develops. This is where the direct effect of the summer weather on the room temperature shows:
At the start of the week, on 21 June, the CPU starts at low load with a completely relaxed operating temperature of around 37 to 38 °C. Even the load spikes barely push it above 40 °C.
But as the days go by, the rooms heat up noticeably. Day after day, the "baseline" (the idle temperature) relentlessly climbs:
- On 23 June the minimum is already closer to 38 to 39 °C.
- On 25 and 26 June the idle temperature is already scratching at the 40-degree mark.
Saturday: the great data vacuum
Take a close look at 27 June (Saturday) and you'll notice a completely unnatural, perfectly straight line in the CPU charts. Neither the CPU load nor the temperature moves even a millimetre. A perfect, magical system setup? If only!
The explanation is much simpler: all of these readings are collected and visualised by my Home Assistant VM. That Saturday I changed a setting on exactly that VM and then simply forgot to start it again.
Since the monitoring instance was offline all Saturday, there were simply no readings. For that period, the dashboard just bridged the gap in the CPU graph with a straight line between the last value before shutdown and the first value after startup. A classic maintenance-window fail!
The hard drives: between spin-down and summer heat
A look at the hard drives also proves that the server room got warmer day by day. Here's the temperature curve of Disk 2 as an example:
The first thing you notice is all the empty gaps in the graph. Don't worry, that's not data loss, it's an intentional feature: when the HDDs aren't needed for a while, they put themselves into standby (spin-down). In that state the drives are asleep, use hardly any power – and logically don't report any temperatures to the monitoring either.
But as soon as the drive is woken up and spins up, the summer heat strikes mercilessly:
- At the start of the week (Sunday to Tuesday) the HDD climbed to a perfectly normal 45 °C when active.
- Towards the end of the week (Friday to Sunday) a spin-up was enough to push the drive to 49 °C, almost 50 °C!
Since hard drives generate their own heat while running, you can really see here how much harder cooling becomes when the air in the room is already heated up.
The furious Sunday finale: fleeing the heat onto the server
On Sunday, 28 June, the Home Assistant VM was running again, and logging resumed right at the moment the curves shot up steeply. And there's a pretty logical reason for that:
It was unbearably hot outside. Instead of roasting in the blazing sun, my family and friends apparently preferred to spend the day indoors in the cool shade – and made heavy use of my server and its services at the same time!
That's where the worst combination of the week came together: the rooms were already heated to the max by the heatwave of the previous days, and at the same time the sudden rush of digital heat refugees kept the hardware absolutely flat out. We can see the result in the week's record peak: the CPU breaks the 46 °C mark, and the busy hard drive scrapes past the magic line at almost 50 °C by a hair's breadth.
Conclusion
The data shows impressively that server monitoring is always partly room-climate and leisure-time monitoring too. When the weather drives people indoors, the hardware glows. Fortunately, for modern systems all of this is still well within the green zone, but it clearly shows where things are headed when summer really cranks it up.