High RAM usage and no process to blame? Check your tmpfs mounts

My computer usually has a couple weeks of runtime between reboots. Being on Linux and using hibernation means I only reboot to update graphical drivers. The routine is as simple as it gets: close all programs one by one so I don’t miss unsaved files. However after shutting down the happy memory hogs – the browser with a hundred tabs and the glutton chat app, I noticed my status bar still showed a memory usage of 10GiB, way above the 2GiB or so I expected¹. I couldn’t find any obvious culprit with htop and, after the reboot, memory usage was back to normal. No more problem, so back to work.

Fast forward a few days, I think again about the problem and want to scratch the itch. I dig a bit and find the answer on the Unix StackExchange: check the space used by tmpfs. I nearly facepalmed: obviously tmpfs – which stores its files in RAM – will use some and not appear in the process list.

So here’s the command to find how much space tmpfs uses:

$ df -h | grep -E "Filesystem|tmpfs"
Filesystem      Size  Used Avail Use% Mounted on
tmpfs            32G  369M   31G   2% /dev/shm
tmpfs            32G  6.7G   25G  22% /tmp
tmpfs           6.3G   92K  6.3G   1% /run/user/1000

which sums up to about 7.1G.

In free and htop, this amount is accounted under the shared memory.

$ free -h
               total        used        free      shared  buff/cache   available
Mem:            62Gi        17Gi        16Gi       7.5Gi        36Gi        45Gi
Swap:             0B          0B          0B

Notice how free reports 17Gi used, while htop reports 9.39G? They both have access to the same info² but they differ in how they interpret it. free’s used is equal to MemTotal - MemAvailable³ while htop sums up the memory used by processes. My status bar – the excellent bumblebee-status – shares the same definition as free.

So… which number is correct? It depends on your use. In my case, I wanted to find a process leaking memory and htop’s “memory used by processes” would have tipped me off. However, if my question was “I want to start a 30GB job, do I have enough headroom?” then free’s available memory gives me what I’m looking for⁴. At least, now I know where to look!