Comment 16 for bug 1899249

So it may be that the write-back-throttling (wbt) for the underlying devices is getting confused about the exact throttle rates are for these devices and somehow getting stuck. It maybe worth experimenting by disabling the throttling and seeing if this gets I/O working again.

For example, to disable wbt for a device /dev/sda use:

echo 0 | sudo tee /sys/block/sda/queue/wbt_lat_usec

and if you need to reset it back to the default:

echo -1 | sudo tee /sys/block/sda/queue/wbt_lat_usec

..use the appropriate block device name for the block devices you have attached. It may even be worth setting the wbt_lat_usec to 0 for all the block devices in your pool as early as possible after boot and see if this helps. My hunch is that wbt is not set correctly for the attached non-spinny devices you have and it is assuming they are reasonably fast when in fact it may be rate limited to be slower than expected because they are attached to relatively slow host.