On 3/26/23 8:49 PM, Christoph Hellwig wrote: > Hi all, > > the current code to offload bio submission into a cgroup-specific helper > thread when sent from the btrfs internal helper threads is a bit ugly. > > This series moves it into btrfs with minimal interference in the core > code. > > I also wonder if the better way to handle this in the long would be to > to allow multiple writeback threads per device and cgroup, which should > remove the need for both the btrfs submission helper workqueue and the > per-cgroup threads. We tried a few different ways to solve this internally, and I'm not against folding it back into btrfs but I'm kinda surprised that nobody else needs it? Is btrfs really the only place using worker threads to submit IO, or are we just the only ones looking for the priority inversions w/resource control in place? -chris
