On Tue, Jun 10, 2025 at 10:02:19AM +0800, Yu Kuai wrote: > Looks like this problem is related to nvme mpath, and it's using > bio based disk IO accounting. I'm not familiar with this driver, > however, can you try the following bpftrace script to check if > start request and end request are balanced? From the log I guess > it's related to mpath error handler, probably requeuing I/O. nvme-mpath is a bio based driver. It is a bit special in that it passed down the bio it recived through to the lower-level request based nvme transport driver, while most remapping drivers always clone a bio. I suspect that is messing up I/O accounting here somehow.
