desktop entry proposal: TerminateSafe=true key

[Patryk Zawadzki]

On Fri, May 7, 2010 at 4:49 PM, Colin Walters <walters at verbum.org> wrote:
> On Fri, May 7, 2010 at 10:38 AM, Lennart Poettering <mzkqt at 0pointer.de> wrote:
>> oom_adj is a way to make the OOM killer smarter. So far only very few
>> apps set that, but we certainly could change that.
> I think asking application authors to patch their apps to use that
> interface is a lot less nice than say asking them to add this key, and
> then if we decided that using the extant oom_adj kernel interface was
> the way to go, the desktop UI could set oom_adj when launching the app
> (after fork, before exec if it's inherited, otherwise take the small
> race condition and set it after invocation).

I'd say "write a gtk loadable module".

>> Hmm, I am not aware that on Linux we have anything like a low-memory
>> signal. Please enlighten me on that!
> We don't right now.

There is no such thing as "low memory". Ideally you should have 100%
of the memory used at all times. For apps, their data, their buffers,
disk buffers etc.

You know you're low on memory when your malloc fails. Failing a malloc
of 500M is not the same as failing a malloc of 4K.

Kernel knows it's low on memory when it badly needs to allocate but
there is not enough memory. It then goes berserk and kills apps.

>> So tell me, why exactly do you want this to be duplicated in userspace?
> We can work out the exact details of how all the components interact
> after the fact.

Asking a userspace app to do anything when allocating 1 byte of memory
is likely to fail is a more polite way of just harvesting it with the
OOM killer. Also, with multiple users logged in, which app should we
ask?

-- 
Patryk Zawadzki