On Tue, Sep 26, 2023 at 8:16 AM Max Kellermann <max.kellermann@xxxxxxxxx> wrote:
>
> On Mon, Sep 25, 2023 at 12:42 PM Ilya Dryomov <idryomov@xxxxxxxxx> wrote:
> > A word of caution about building metrics collectors based on debugfs
> > output: there are no stability guarantees. While the format won't be
> > changed just for the sake of change of course, expect zero effort to
> > preserve backwards compatibility.
>
> Agree, but there's nothing else. We have been using my patch for quite
> some time, and it has been very useful.
>
> Maybe we can discuss promoting these statistics to sysfs/proc? (the
> raw numbers, not the existing aggregates which are useless for any
> practical purpose)
>
> > The latency metrics in particular are sent to the MDS in binary form
> > and are intended to be consumed through commands like "ceph fs top".
> > debugfs stuff is there just for an occasional sneak peek (apart from
> > actual debugging).
>
> I don't know the whole Ceph ecosystem so well, but "ceph" is a command
> that is supposed to run on a Ceph server, and not on a machine that
> mounts a cephfs, right? If that's right, then this command is useless
> for me.
No, "ceph" command (as well as "rbd", "rados", etc) can be run from
anywhere -- it's just a matter of installing a package which is likely
already installed unless you are mounting CephFS manually without using
/sbin/mount.ceph mount helper.
Thanks,
Ilya
