Saw the issue, thanks! Looking forward to try the fix, I have noticed that once the dataset grows more than what the OS cache can handle, we are back to square one as far as stalls go. Tomorrow I will go back to 16k (sorry!) and to 50% tokudb_cache_mem which offered the best results in our situation so far. Mind you, with 8 GB toku cache and 64k once the dataset hit the disk, I have been able to saturate the tokudb_cachetable_pool_threads default value of 80 (the only one of out the 3 that I left to default). Not sure why this happened or what that means actually, but having the pool saturated for one minute when I had the stall just seemed related. Thanks! Rick -- Riccardo Pizzi