Support to single cgroup mountpoint?

[ghedo]

Hi,

currently ulatencyd uses different mountpoints for cgroupfs (/sys/fs/cgroup/cpu, /sys/fs/cgroup/memory, etc...). What about supporting also a single mountpoint for them all (/sys/fs/cgroup)? Do you think it is doable? What kind of issues this may introduce?

I'm asking because there is a on going discussion in Debian about mounting cgroupfs by default in /sys/fs/cgroup and this would break ulatencyd. This is also valid if someone decides to mount cgroupfs in fstab (see #643580).

[poelzi]

it is possible with configs changes, but not very practical. the problem is, that different cgoup subsystem support different hierarchy levels, blockio for example only supports one level. Mounting all subsystem into /sys/fs/cgroup, which i tried is more problematic then different mountpoints. You will not be able to put programs in different groups for different subsystems etc.

[ghedo]

May this be supported as a fallback? It would be nice if ulatencyd worked in this situation (maybe after printing a warning) without aborting. The default behaviour would remain the same.

[poelzi]

yes, that could be done, but requires duplicaton of rules. where is the discussion going on, so i could throw my 2cents at it :-)

[ghedo]

The discussion about mounting cgroup by default is here (now that you ask it seems dead...), while the report about ulatencyd not working with it is linked above.

Even if the discussion about mounting cgroup by default goes nowhere, someone still do that using fstab along with ulatencyd, should I simply suggest to stop doing it?

[rickysarraf]

Any update on this bug?

The discussion pointed by AlexBio is for enabling cgroup automount at mount point /sys/fs/cgroup

The current approach that ulatency takes makes it incompatible with the rest. At least, Linux Containers, it expects cgroup to be mounted at /sys/fs/cgroup , making it difficult to use lxc and ulatency at the same time.
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=645995

[poelzi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Mon, 5 Dec 2011 11:58:55 -0800
Ritesh Raj Sarraf
[email protected]
wrote:

Any update on this bug?

The discussion pointed by AlexBio is for enabling cgroup automount at
mount point /sys/fs/cgroup

The current approach that ulatency takes makes it incompatible with
the rest. At least, Linux Containers, it expects cgroup to be mounted
at /sys/fs/cgroup , making it difficult to use lxc and ulatency at
the same time. http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=645995

ulatency can be adjusted to a single mountpoint, but as said many
times, it will suck and make it harder und more complicated to use. CPU
subsystem is totally broken when Realtime Cgroups is enabled and you
need constant fixing on that, making it harder with unified mountpoints.
You can not even create more then one level of in blkio cgroup which
will make the cpu configuration useless.
Single mount point of all cgroup subsystems IS A BAD IDEA. You can
configure it to do so, but it will not help you.

kind regards
Daniel
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)

iEYEARECAAYFAk7d9sUACgkQy/mkIQp7AD3sjQCeI6pB8Mu9C0IiSMgiYmydpKYs
9twAn327HoPDRBji+k+iaKW0t6yPjEOL
=2Qhn
-----END PGP SIGNATURE-----

[rickysarraf]

On 12/06/2011 04:35 PM, Daniel Poelzleithner wrote

ulatency can be adjusted to a single mountpoint, but as said many
times, it will suck and make it harder und more complicated to use. CPU
subsystem is totally broken when Realtime Cgroups is enabled and you
need constant fixing on that, making it harder with unified mountpoints.
You can not even create more then one level of in blkio cgroup which
will make the cpu configuration useless.
Single mount point of all cgroup subsystems IS A BAD IDEA. You can
configure it to do so, but it will not help you.

kind regards
Daniel

Thanks Daniel. I have taken your comments forward to the concerned folks.

Ritesh Raj Sarraf
RESEARCHUT - http://www.researchut.com
"Necessity is the mother of invention."