[PATCH] tst-malloc-thread-exit: run less aggressively

[Florian Weimer]

On 01/19/2016 09:04 PM, Florian Weimer wrote:
> On 01/19/2016 08:58 PM, Chris Metcalf wrote:
>> On a 64-bit machine, the previous code would run up to
>> 5 x 8 x NPROCS threads, or 2,880 threads on a 72-core TILE-Gx.
>> But since typically userspace runs with an RLIMIT_NPROC value
>> of 1024, the test will fail on such a machine.  Instead, bound
>> the outer thread count to 200, rather than 8 x NPROCS, so that
>> the total number of threads created is max 1000.  This may still
>> be a little risky when running in a "make -j check" context but
>> should at least pass when run separately to confirm a FAIL.
> 
> I posted a patch due to similar issues:
> 
>   <https://sourceware.org/ml/libc-alpha/2015-12/msg00537.html>
> 
> This change will bring down the run time considerably.
> 
>> Can I push this for 2.23?  It fixes a test failure for tilegx.
> 
> Not just there.
> 
> I wasn't aware of the thread count limit.  I can post a patch tomorrow
> which uses mallopt to reduce the arena count, then we should be able to
> run this test with just 8 threads (but I will have to verify it still
> triggers the original failure).

I haven't finished the mallopt approach.  It seems that my mental model
how the test works is wrong.  Two threads with eight arenas appear to
trigger the issue reliably, but I want to test on a larger machine as
well, to make sure the test is any good there.  The problem there is
that I wasn't

I'm attaching what I'm using for testing.  Warning: It contains a revert
of the bug fix.

Florian

-------------- next part --------------
A non-text attachment was scrubbed...
Name: thread-exit-test.patch
Type: text/x-patch
Size: 1220 bytes
Desc: not available
URL: <http://sourceware.org/pipermail/libc-alpha/attachments/20160122/1d56d75e/attachment.bin>