Re: [PATCH] mm/vmscan: add vm_swappiness configuration knobs

[Matthew Wilcox]

On Thu, Mar 12, 2020 at 12:48:22PM +0000, Ivan Teterevkov wrote:
> On Wed, 11 Mar 2020, David Rientjes wrote:
> > On Wed, 11 Mar 2020, Ivan Teterevkov wrote:
> > > This patch adds a couple of knobs:
> > >
> > > - The configuration option (CONFIG_VM_SWAPPINESS).
> > > - The command line parameter (vm_swappiness).
> > >
> > > The default value is preserved, but now defined by CONFIG_VM_SWAPPINESS.
> > >
> > > Historically, the default swappiness is set to the well-known value 60,
> > > and this works well for the majority of cases. The vm_swappiness is also
> > > exposed as the kernel parameter that can be changed at runtime too, e.g.
> > > with sysctl.
> > >
> > > This approach might not suit well some configurations, e.g. systemd-based
> > > distros, where systemd is put in charge of the cgroup controllers,
> > > including the memory one. In such cases, the default swappiness 60
> > > is copied across the cgroup subtrees early at startup, when systemd
> > > is arranging the slices for its services, before the sysctl.conf
> > > or tmpfiles.d/*.conf changes are applied.
> > 
> > Seems like something that can be fully handled by an initscript that would
> > set the sysctl and then iterate the memcg hierarchy propagating the
> > non-default value.  I don't think that's too much of an ask if userspace
> > wants to manipulate the swappiness value.
> > 
> 
> This is exactly what I'm trying to avoid: in some distros there is no way
> to tackle the configuration early enough, e.g. in systemd-based systems
> the systemd is the process that starts first and arranges memcg in a way
> it's configured, but unfortunately, it doesn't offer the swappiness knob.

This sounds like a systemd problem.  Have you talked to the systemd
people about fixing it in systemd?