The xmr-stak ebuild creates a sandbox for xmr-stak using systyemd. In the default configuration the xmr-stak.service unit file creates an error (amdgpu,nvidia). When the sandboxing features are disabled, and xmr-stak is instead given it's own user, then it's able to initialize the cuda/opencl drivers. Upon repeat invocations(even after clearing and reloading the unit file) xmr-stak.service will continue to operate properly until the nvidia driver is removed and reinserted. Reproducible: Always Steps to Reproduce: 1.systemctl start xmr-stak 2.see failure 3.patch systemd unit to give xmr-stak it's own user/remove extra sandboxing 4.normal behaviour Actual Results: Sep 09 00:35:24 undyne xmr-stak[15634]: WARNING: NVIDIA Unable to query number of CUDA devices! Sep 09 00:35:24 undyne xmr-stak[15634]: WARNING: NVIDIA no device found Sep 09 00:35:24 undyne xmr-stak[15634]: [2018-09-09 00:35:24] : WARNING: backend NVIDIA disabled. Expected Results: Sep 09 00:31:04 undyne xmr-stak[3471]: [2018-09-09 00:31:04] : Starting NVIDIA GPU thread 0, no affinity. I've tried individually individually disabling each of the options in the unit file and I can't seem to narrow it down to a single option that's causing the failure. The only thing that seems consistent is replacing DynamicUsers with User=foo is required for this to work
Created attachment 546320 [details, diff] Patching fixing unit file
Created attachment 546322 [details] xmr-stak cpu config
Created attachment 546324 [details] xmr-stak gpu config
Created attachment 546326 [details] xmr-stak main config
Created attachment 546328 [details] xmr-stak pool config Demonstration purposes, you would have to enter your address before this file would work
removed from portage