[systemd-devel] [PATCH 1/3] Enable systemd to set/show limit of memory+Swap usage

[Lennart Poettering]

On Tue, 17.09.13 17:36, Lennart Poettering (lennart at poettering.net) wrote:

> On Tue, 17.09.13 14:21, Chen Hanxiao (chenhanxiao at cn.fujitsu.com) wrote:
> 
> > From: Chen Hanxiao <chenhanxiao at cn.fujitsu.com>
> > 
> > @@ -89,6 +89,7 @@ $1.CPUAccounting,                config_parse_bool,                  0,
> >  $1.CPUShares,                    config_parse_cpu_shares,            0,                             offsetof($1, cgroup_context)
> >  $1.MemoryAccounting,             config_parse_bool,                  0,                             offsetof($1, cgroup_context.memory_accounting)
> >  $1.MemoryLimit,                  config_parse_memory_limit,          0,                             offsetof($1, cgroup_context)
> > +$1.MemorySWLimit,                config_parse_memory_limit,          0,                             offsetof($1, cgroup_context)
> >  $1.MemorySoftLimit,              config_parse_memory_limit,          0,                             offsetof($1, cgroup_context)
> >  $1.DeviceAllow,                  config_parse_device_allow,          0,                             offsetof($1, cgroup_context)
> >  $1.DevicePolicy,                 config_parse_device_policy,         0,                             offsetof($1, cgroup_context.device_policy)
> 
> OK, so here's another idea: I have the strong suspicion that people are
> much more likely using the new limit that includes the swap than the
> current MemoryLimit= that doesn't.
> 
> Hence, to make this simpler, I'd propose to simply swap things around:
> 
> MemoryLimit= would start writing to memory.memsw.limit_in_bytes. And a
> new MemoryRAMLimit= would controler the original memory.limit_in_bytes?
> 
> This shifts things around a bit but I think it would be much nicer to
> use?

OK, so I talked to Tejun here at LinuxCon and he said that we probably
should not expose memory.memsw.limit_in_bytes for now since it's likely
to change or go away soon. In fact, the only memory attribute we should
expose for now is MemoryLimit= accorrding to him, and it should do what
it already does.

Also, he said that memory.use_hierarchy should be unconditionally set by
systemd for all cgroups systemd creates.

Lennart

-- 
Lennart Poettering - Red Hat, Inc.