[squeak-dev] [Vm-dev] testOutOfMemorySignalExtreme

Marcel Taeumel marcel.taeumel at hpi.de
Wed Aug 4 07:22:06 UTC 2021


Hi all --

If we really want to test an out-of-memory event, we might want to implement a configurable boundary in the VM to accommodate today's operating systems as Tobias explained.

In the past, this test relied on an implementation detail in the OS that is no longer valid. Such tests are never a good idea. Instead, let's replace this test with something like:

testOutOfMemorySignalExtreme

   self setMemoryLimit: 2 * 1024 * 1024 * 1024.
   self should: [self useUpAllMemory] raise: OutOfMemory.

At the moment, such a memory limit is currently not configurable in recent VMs. If we had such a limit, one could set it to match a machine's physical memory -- or 80% of it -- to not stress SSDs too much because of paging.

Of course, such a limit could be "undefined" to realize the current situation. And for backwards compatibility.

Best,
Marcel
Am 03.08.2021 22:19:14 schrieb Tobias Pape <das.linux at gmx.de>:
Hi


> On 3. Aug 2021, at 20:05, Eliot Miranda wrote:
>
> Hi All,
>
> something needs to be done about this test. Whenever I run it on macos it never runs out of memory before the system starts complaining. I've just been presented with a dialog telling me that the system is out of application memory and something needs to quit, while the Squeak process is 236 Gb big. A test that locks up the system isn't very useful.

macOS uses the dynamic_pager(8) to handle swap.
It tries very very hard to satisfy your ram demands.
just out of curiosity, do a

ls -hal /var/vm

and see all the swap files it happily creates when you ask for more ram than you have.

The test really only makes sense if the OS does not pretend to have an infinite amount of random access bits.
Thus probably only sensible on linux these days (it presume even iOS has the dynamic pager…)

(Note: it seems dynamic_pager got stripped down over releases, and does not do much anymore;
functionality still exists but is less configurable and problably in-kernel only.
Main point stands tho)

Best regards
-Tobias



> _,,,^..^,,,_
> best, Eliot



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squeakfoundation.org/pipermail/squeak-dev/attachments/20210804/d7f70c81/attachment.html>


More information about the Squeak-dev mailing list