I learned about the kernel watchdog timer when I ran my network appliances on hardware from Soekris Engineering. The appliances should be running unless one has specifically turned them off but as the that’s not the reality of the world. A watchdog timer is something that has to be frobbed, I’ve also seen petted periodically or the kernel will assume that all of userland is borked. When userland is borked this way, the kernel will reset, hoping that a reset will kickstart enough of the userland infrastructure programs to put the system into a workable state. On Soekris, you could program the kernel watchdog to timeout of 60s. On the Pi, it looks like the maximum is 15s. Raspberry Pi’s are weird. For small jobs, the best way to run them is on SD cards.But an SD card in a Pi will die from write exhaustion after running for something between a year and two years. This is my experience based on using SanDisk 8G SD cards without any consideration for write exhaustion. You can mitigate this and lengthen the lifetime in many ways:
- Purchase a 16G card where you have an 8G need.
- Purchase a Pi with more RAM than you need and perform write heavy task to a memory backed filesystem or ramdisk.
- Do all of your logging via syslog and push all your logs onto a remote log server.
On the Pi devices that I use as cameras, I write the picture / movie output to a RAM disk. I have also noticed that the older SD devices I was buying in the past were more susceptible to this problem that the devices that I buy lately. This is probably a biased observation.