- Annoyance
I have "reliably" caused "Out of memory" crashes under the following condition:
- Pin dies without direct memory access to some dockers
- Put those dockers under high load (like running 3 simultaneous Handbrake dockers transcoding H265)
- Leave about 1%-2% of memory free (about 80% occupied by VM / dockers, 18-19% by RAM cache)
Under this condition, after about 10-15 minutes, some processes are automatically killed by unRAID with Out of Memory error. That's unexpected in this scenario because there's constantly 1%-2% memory totally free (manually monitored) + 18% "buffer" (RAM cache).
Spreading the pinned cores across all 4 dies do not lead to this Out of Memory error so I suspect this has something to do with Threadripper 2 optimisation so probably a kernel problem and not actually unRAID bug.
At this point, it's more an annoyance for me but I'm sure there's someone out there who might be caught off guard.