@phnt i'm running on a 3TB slab with a 425GB lvmcache. And yeah, it chokes due to what I believe are arbitary IOPS or storage bandwidth limits.

What I end up seeing is a complete drop-out of the storage for upwards of 15 seconds and then all the queued I/O flushes at once. rinse and repeat.

Better yet, if you don't use the storage for like a whole day, the iowait is basically nil. Then out of the blue, the issue comes back!