Message-ID: <3dd64de1-8762-4870-84b3-fb2c5234f736@redhat.com>
Date: Tue, 9 Jul 2024 20:23:28 -0400
From: Waiman Long <longman@...hat.com>
To: Roman Gushchin <roman.gushchin@...ux.dev>
Cc: Tejun Heo <tj@...nel.org>, Zefan Li <lizefan.x@...edance.com>,
 Johannes Weiner <hannes@...xchg.org>, Jonathan Corbet <corbet@....net>,
 cgroups@...r.kernel.org, linux-doc@...r.kernel.org,
 linux-kernel@...r.kernel.org
Subject: Re: [PATCH-cgroup v2] cgroup: Show # of subsystem CSSes in root
 cgroup.stat

On 7/9/24 19:08, Roman Gushchin wrote:
> On Tue, Jul 09, 2024 at 09:28:14AM -0400, Waiman Long wrote:
>> The /proc/cgroups file shows the number of cgroups for each of the
>> subsystems.  With cgroup v1, the number of CSSes is the same as the
>> number of cgroups. That is not the case anymore with cgroup v2. The
>> /proc/cgroups file cannot show the actual number of CSSes for the
>> subsystems that are bound to cgroup v2.
>>
>> So if a v2 cgroup subsystem is leaking cgroups (usually memory cgroup),
>> we can't tell by looking at /proc/cgroups which cgroup subsystems may be
>> responsible.  This patch adds CSS counts in the cgroup_subsys structure
>> to keep track of the number of CSSes for each of the cgroup subsystems.
>>
>> As cgroup v2 had deprecated the use of /proc/cgroups, the root
>> cgroup.stat file is extended to show the number of outstanding CSSes
>> associated with all the non-inhibited cgroup subsystems that have been
>> bound to cgroup v2.  This will help us pinpoint which subsystems may be
>> responsible for the increasing number of dying (nr_dying_descendants)
>> cgroups.
>>
>> The cgroup-v2.rst file is updated to discuss this new behavior.
>>
>> With this patch applied, a sample output from root cgroup.stat file
>> was shown below.
>>
>> 	nr_descendants 53
>> 	nr_dying_descendants 34
>> 	nr_cpuset 1
>> 	nr_cpu 40
>> 	nr_io 40
>> 	nr_memory 87
>> 	nr_perf_event 54
>> 	nr_hugetlb 1
>> 	nr_pids 53
>> 	nr_rdma 1
>> 	nr_misc 1
>>
>> In this particular case, it can be seen that memory cgroup is the most
>> likely culprit for causing the 34 dying cgroups.
>>
>> Signed-off-by: Waiman Long <longman@...hat.com>
>> ---
>>   Documentation/admin-guide/cgroup-v2.rst | 10 ++++++++--
>>   include/linux/cgroup-defs.h             |  3 +++
>>   kernel/cgroup/cgroup.c                  | 19 +++++++++++++++++++
>>   3 files changed, 30 insertions(+), 2 deletions(-)
>>
>> diff --git a/Documentation/admin-guide/cgroup-v2.rst b/Documentation/admin-guide/cgroup-v2.rst
>> index 52763d6b2919..65af2f30196f 100644
>> --- a/Documentation/admin-guide/cgroup-v2.rst
>> +++ b/Documentation/admin-guide/cgroup-v2.rst
>> @@ -981,6 +981,12 @@ All cgroup core files are prefixed with "cgroup."
>>   		A dying cgroup can consume system resources not exceeding
>>   		limits, which were active at the moment of cgroup deletion.
>>   
>> +	  nr_<cgroup_subsys>
>> +		Total number of cgroups associated with that cgroup
>> +		subsystem, e.g. cpuset or memory.  These cgroup counts
>> +		will only be shown in the root cgroup and for subsystems
>> +		bound to cgroup v2.
>> +
>>     cgroup.freeze
>>   	A read-write single value file which exists on non-root cgroups.
>>   	Allowed values are "0" and "1". The default is "0".
>> @@ -2930,8 +2936,8 @@ Deprecated v1 Core Features
>>   
>>   - "cgroup.clone_children" is removed.
>>   
>> -- /proc/cgroups is meaningless for v2.  Use "cgroup.controllers" file
>> -  at the root instead.
>> +- /proc/cgroups is meaningless for v2.  Use "cgroup.controllers" or
>> +  "cgroup.stat" files at the root instead.
>>   
>>   
>>   Issues with v1 and Rationales for v2
>> diff --git a/include/linux/cgroup-defs.h b/include/linux/cgroup-defs.h
>> index b36690ca0d3f..522ab77f0406 100644
>> --- a/include/linux/cgroup-defs.h
>> +++ b/include/linux/cgroup-defs.h
>> @@ -776,6 +776,9 @@ struct cgroup_subsys {
>>   	 * specifies the mask of subsystems that this one depends on.
>>   	 */
>>   	unsigned int depends_on;
>> +
>> +	/* Number of CSSes, used only for /proc/cgroups */
>> +	atomic_t nr_csses;
> I believe it should be doable without atomics because most of css
> operations are already synchronized using the cgroup mutex.
css_create() was protected under cgroup_mutex, but I don't believe 
css_free_rwork_fn() is. It is called from the kworker. So atomic_t is 
still needed.
>
> Other than that, I believe that this information is useful. Maybe
> it can be retrieved using drgn/bpf iterator, but adding this functionality
> to the kernel makes it easier to retrieve and the overhead is modest.
>
> Also, if you add it to the cgroupfs, why not make it fully hierarchical
> as existing entries in cgroup.stat. And if not, I'd agree with Johannes
> that it looks like the debugfs material.

To make it hierarchical, I would have to store a nr_descendants and 
nr_dying_descendants in each css, just like the corresponding ones in 
cgroup. I think it is doable, but the patch will be much more complex.

Cheers,
Longman

>
> Thanks!
>

Powered by blists - more mailing lists