Hello, Weiping. On Fri, Mar 27, 2020 at 12:27:11AM +0800, Weiping Zhang wrote: > I really appreciate that if you help review blk-iotrack.c, or just > drop io.iotrakc.stat > and append these statistics to the io.stat? I think these metrics is usefull, So, the problem is that you can get the same exact and easily more information using bpf. There definitely are benefits to baking in some stastics in terms of overhead and accessbility but I'm not sure what's being proposed is generic and/or flexible enough to bake into the interface at this point. Something which can be immediately useful would be cgroup-aware bpf progs which expose these statistics. Can you please take a look at the followings? https://github.com/iovisor/bcc/blob/master/tools/biolatency.py https://github.com/iovisor/bcc/blob/master/tools/biolatency_example.txt They aren't cgroup aware but can be made so and can provide a lot more detailed statistics than something we can hardcode into the kernel. Thanks. -- tejun
