Hi Michael,
El 17/03/21 a las 15:01, Michael Biebl escribió:
> Hi Santiago
>
> On Fri, 12 Mar 2021 21:33:05 +0100 Santiago Ruano =?iso-8859-
> 1?Q?Rinc=F3n?= <santiagorr@riseup.net> wrote:
>
> > Dear maintainers of the packages that would be removed by the removal
> of
> > libcgroup,
>
>
> Checking reverse dependencies...
> # Broken Depends:
> cinder: cinder-common → cgroup-tools
> clsync: clsync → libcgroup1
> condor: htcondor → libcgroup1
> mininet: mininet → cgroup-tools
> nova: nova-compute → cgroup-tools
> vzctl: vzctl [amd64 i386] → libcgroup1
>
> # Broken Build-Depends:
> clsync: libcgroup-dev
> condor: libcgroup-dev
> vzctl: libcgroup-dev
>
>
>
> > I am taking a look at the above mentioned new version of libcgroup,
> that
> > partially supports cgroup2. Would it make sense for your packages to
> > have in Debian that version of libcgroup (the current git HEAD)?
>
> While I appreciate the effort, I feel a bit uneasy about shipping
> something half-baked like this. Does such an updated libcgroup package
> actually ensure, that rdeps continue to work?
Yeah, answering that was actually the main purpose of my question!
>
> Maybe the packages above need to check, whether they actually need the
> libcgroup / cgroup-tools dependency.
Yes, indeed. I have been only able to take a look at mininet, that I use
personally a little bit. In a basic use as mine, mininet works (so I
could drop the dependency).
But it is broken if you need to do stuff that relies on cgroups:
e.g. running `python3 /usr/share/doc/mininet/examples/limit.py`
fails:
Traceback (most recent call last):
File "/usr/share/doc/mininet/examples/limit.py", line 61, in <module>
limit()
File "/usr/share/doc/mininet/examples/limit.py", line 37, in limit
net = Mininet( topo=myTopo, intf=intf, host=host, waitConnected=True )
File "/usr/lib/python3/dist-packages/mininet/net.py", line 178, in __init__
self.build()
File "/usr/lib/python3/dist-packages/mininet/net.py", line 508, in build
self.buildFromTopo( self.topo )
File "/usr/lib/python3/dist-packages/mininet/net.py", line 479, in buildFromTopo
self.addHost( hostName, **topo.nodeInfo( hostName ) )
File "/usr/lib/python3/dist-packages/mininet/net.py", line 232, in addHost
h = cls( name, **defaults )
File "/usr/lib/python3/dist-packages/mininet/util.py", line 587, in customized
return cls( *args, **kwargs )
File "/usr/lib/python3/dist-packages/mininet/node.py", line 692, in __init__
CPULimitedHost.init()
File "/usr/lib/python3/dist-packages/mininet/node.py", line 871, in init
mountCgroups()
File "/usr/lib/python3/dist-packages/mininet/util.py", line 548, in mountCgroups
raise Exception( "cgroups not mounted on " + cgdir )
Exception: cgroups not mounted on /sys/fs/cgroup
mininet code that relies on cgroupv1, e.g.:
https://sources.debian.org/src/mininet/2.3.0-1/mininet/node.py/#L695
https://sources.debian.org/src/mininet/2.3.0-1/mininet/util.py/#L541
should have to be patched to be able to run with cgroupv2. A priori, the
upgraded cgroup-tools would make that easier.
I have no idea about cinder and nova, I let the maintainers give their input.
>
> If this was broken for such a long time with noone noticing, maybe
> dropping that dependency is the better alternative. At least the
> cgroup-tools dependency looks like something that could be dropped.
Maybe. It is maybe difficult to notice since some tools don't have a
significant number of users that run testing or unstable (that's my case
with mininet).
Cheers,
-- Santiago
Attachment:
signature.asc
Description: PGP signature