Messages in this thread | | | Date | Thu, 17 May 2012 09:01:31 +0900 | From | KAMEZAWA Hiroyuki <> | Subject | Re: [patch 6/6] mm: memcg: print statistics from live counters |
| |
(2012/05/17 8:01), Andrew Morton wrote:
> On Mon, 14 May 2012 20:00:51 +0200 > Johannes Weiner <hannes@cmpxchg.org> wrote: > >> Directly print statistics and event counters instead of going through >> an intermediate accumulation stage into a separate array, which used >> to require defining statistic items in more than one place. >> >> ... >> >> -static const char *memcg_stat_strings[NR_MCS_STAT] = { >> - "cache", >> - "rss", >> - "mapped_file", > > Bah humbug, who went and called this mapped_file? > > This stat is derived from MEM_CGROUP_STAT_FILE_MAPPED. But if we > rename MEM_CGROUP_STAT_FILE_MAPPED to MEM_CGROUP_STAT_MAPPED_FILE then > we also need to rename the non-memcg NR_FILE_MAPPED. And we can't > change the text to "file_mapped" because it's ABI. >
Sorry..
>> - "mlock", >> - "swap", > > And "swap" is derived from MEM_CGROUP_STAT_SWAPOUT. We could rename > that to MEM_CGROUP_STAT_SWAP without trouble. >
Yes.
> But both are poor names. There are two concepts here: a) swapout > events (ie: swap writeout initiation) and b) swapspace usage. Type a) > only ever counts up, whereas type b) counts up and down. > > MEM_CGROUP_STAT_SWAPOUT is actually of type b), but "swapout" is a > misleading term, because it refers to type a) events. >
I'll prepare a patch.
> And the human-displayed "swap" is useless because it can refer to > either type a) or type b) events. These should be called "swapped" and > MEM_CGROUP_STAT_SWAPPED. But we can't change the userspace interface. > > argh, I hate you all! >
Hm...sorry. I(fujitsu) am now considering to add meminfo for memcg...,
add an option to override /proc/meminfo if a task is in container or meminfo file somewhere. (Now, we cannot trust /usr/bin/free, /usr/bin/top etc...in a container.)
so...I think usual user experience will be better because of the same format with meminfo.
Thanks, -Kame
| |