lkml.org 
[lkml]   [2012]   [Aug]   [22]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    From
    SubjectRe: [PATCH v2 09/11] memcg: propagate kmem limiting information to children
    Date
    On Wed, Aug 22 2012, Glauber Costa wrote:

    >>>>
    >>>> I am fine with either, I just need a clear sign from you guys so I don't
    >>>> keep deimplementing and reimplementing this forever.
    >>>
    >>> I would be for make it simple now and go with additional features later
    >>> when there is a demand for them. Maybe we will have runtimg switch for
    >>> user memory accounting as well one day.
    >>>
    >>> But let's see what others think?
    >>
    >> In my use case memcg will either be disable or (enabled and kmem
    >> limiting enabled).
    >>
    >> I'm not sure I follow the discussion about history. Are we saying that
    >> once a kmem limit is set then kmem will be accounted/charged to memcg.
    >> Is this discussion about the static branches/etc that are autotuned the
    >> first time is enabled?
    >
    > No, the question is about when you unlimit a former kmem-limited memcg.
    >
    >> The first time its set there parts of the system
    >> will be adjusted in such a way that may impose a performance overhead
    >> (static branches, etc). Thereafter the performance cannot be regained
    >> without a reboot. This makes sense to me. Are we saying that
    >> kmem.limit_in_bytes will have three states?
    >
    > It is not about performance, about interface.
    >
    > Michal says that once a particular memcg was kmem-limited, it will keep
    > accounting pages, even if you make it unlimited. The limits won't be
    > enforced, for sure - there is no limit, but pages will still be accounted.
    >
    > This simplifies the code galore, but I worry about the interface: A
    > person looking at the current status of the files only, without
    > knowledge of past history, can't tell if allocations will be tracked or not.

    In the current patch set we've conflating enabling kmem accounting with
    the kmem limit value (RESOURCE_MAX=disabled, all_other_values=enabled).

    I see no problem with simpling the kernel code with the requirement that
    once a particular memcg enables kmem accounting that it cannot be
    disabled for that memcg.

    The only question is the user space interface. Two options spring to
    mind:
    a) Close to current code. Once kmem.limit_in_bytes is set to
    non-RESOURCE_MAX, then kmem accounting is enabled and cannot be
    disabled. Therefore the limit cannot be set to RESOURCE_MAX
    thereafter. The largest value would be something like
    RESOURCE_MAX-PAGE_SIZE. An admin wondering if kmem is enabled only
    has to cat kmem.limit_in_bytes - if it's less than RESOURCE_MAX, then
    kmem is enabled.

    b) Or, if we could introduce a separate sticky kmem.enabled file. Once
    set it could not be unset. Kmem accounting would only be enabled if
    kmem.enabled=1.

    I think (b) is clearer.


    \
     
     \ /
      Last update: 2012-08-23 01:41    [W:2.193 / U:1.228 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site