I'm chasing a theory that the kernel watchdog on one my ARM machines resets the device during IBD, often during heavier I/O like pruning or cache flushes. Is that even plausible? And is there anything we should do to put said watchdog at ease? http://forum.khadas.com/t/how-to-debug-spontaneous-resets/2624
provoostenator: not exactly on topic for this channel, but I can comment from my experience that I had ARM devices overheat with CPU load due to inadequate heatsinking, causing such problems.
warran: thanks. I'll let the folks on that forum help with the device itself. The part that's more relevant here is that these resets often leave the chain in an unrecoverable state, at least with pruning enabled. And maybe there's a way to shut down gracefully or not use max resources that we can recommend / do by default on these devices.
provoostenator: totally plausable. one of my older odroids does it reliably. adding sleeps to the sync OR putting a fan on top of it, fixes it.
It's really hard to deal with unreliable hardware, nothing in software is written to handle mystery memory corruption or similar, so all bets are off.
I did add a heatsink and fan btw, but I wouldn't be surprised if Kernel / OS support for that is misconfigured.
The heatsink + fan that came with one of my ODroids a few years ago was inadequate to cool the device and keep it reliable while under full CPU load on the four cores.
I made good experience with the Odroid HC1 (passive / heatsink) and the XU4 with heatsink. Testing both since 3 months.