I've read log files in detail and found relation:
after one or more GPUs stops giving shares, "total table" stops appearing further. In below example GPU3 and GPU4 are lost, so the blue table is not shown anymore, but rig continues hashing without two GPUs until manual restart. I'm sure you can catch totals disappearing in code and react as GPU watchdog shoud. I could write external tool for log monitoring, but it's a hack.

[2019-08-27 23:14:49] Pool xmr-eu1.nanopool.org share accepted. (GPU1) (a:47561 r:853) (128 ms)
[2019-08-27 23:14:53] Pool xmr-eu1.nanopool.org share accepted. (GPU0) (a:47562 r:853) (188 ms)
[2019-08-27 23:14:56] Pool xmr-eu1.nanopool.org share accepted. (GPU2) (a:47563 r:853) (142 ms)
[2019-08-27 23:14:58] Pool xmr-eu1.nanopool.org received new job. (job_id: 9115)
[2019-08-27 23:15:11] Pool xmr-eu1.nanopool.org share accepted. (GPU6) (a:47564 r:853) (158 ms)
[2019-08-27 23:15:23] Pool xmr-eu1.nanopool.org received new job. (job_id: 9116)
[2019-08-27 23:15:46] Pool xmr-eu1.nanopool.org share accepted. (GPU6) (a:47565 r:853) (135 ms)
[2019-08-27 23:15:47] Pool xmr-eu1.nanopool.org share accepted. (GPU0) (a:47566 r:853) (159 ms)