diff mbox

nptl/tst-robust-fork: Increase timeout to 30 seconds.

Message ID a062e92a-b368-90e2-9a70-bdb810e6b484@redhat.com
State New
Headers show

Commit Message

Carlos O'Donell May 18, 2017, 7:42 p.m. UTC 
On 05/18/2017 03:28 PM, Tulio Magno Quites Machado Filho wrote:
> Carlos O'Donell <carlos@redhat.com> writes:
> 
>> The default test timeout is 20 seconds.
>>
>> On certain x86_64 hardware I can see tst-robust-fork test take 22 seconds on average.
>>
>> This results in spurious failures.
>>
>> I suggest we increase the timeout to 25 seconds and add a note about this.
>>
>> If nobody objects I'll check this in at some point next week when I get a
>> free minute.
>>
>> I don't know if it's particularly useful to list the kernel version or
>> hardware under which the slow timing came from i.e. linux 4.10 on an older
>> i5-4690K.
>>
>> 2017-05-18  Carlos O'Donell  <carlos@redhat.com>
>>
>> 	* nptl/tst-robust-fork.c (TIMEOUT): Define.
> 
> I agree with this change.
> 
> I have already proposed another patch increasing it to 35s [1].
> IMO, 25 is too close to the time we both measured.
> 
> By the way, it isn't restricted to x86_64.  I've seen this on powerpc as well.
> 
> [1] https://patchwork.sourceware.org/patch/19174/
 
I agree with you.

I disagree with Florian on the basis that we need to do
_everything_ possible to have `make check` just work (tm)
so developers don't have to set TIMEOUTFACTOR for average
developer machines.

I'm OK with 35s. I'll raise it to 35s and simplify the
comment a bit.

v2
- Raise to 35s.

OK with this?

2017-05-18  Tulio Magno Quites Machado Filho  <tuliom@linux.vnet.ibm.com>
	    Carlos O'Donell  <carlos@redhat.com>

	* nptl/tst-robust-fork.c (TIMEOUT): Set to 35.

Diff from v1:

---

Comments

Florian Weimer May 18, 2017, 7:53 p.m. UTC | #1 
On 05/18/2017 09:42 PM, Carlos O'Donell wrote:
> I disagree with Florian on the basis that we need to do
> _everything_ possible to have `make check` just work (tm)
> so developers don't have to set TIMEOUTFACTOR for average
> developer machines.
> 
> I'm OK with 35s. I'll raise it to 35s and simplify the
> comment a bit.
> 
> v2
> - Raise to 35s.
> 
> OK with this?

I meant to come back to this.  I was wrong, raising the timeout is fine.
 I personally did this for some other tests, after all.

Thanks,
Florian
Tulio Magno Quites Machado Filho May 18, 2017, 8:05 p.m. UTC | #2 
Carlos O'Donell <carlos@redhat.com> writes:

> I agree with you.
>
> I disagree with Florian on the basis that we need to do
> _everything_ possible to have `make check` just work (tm)
> so developers don't have to set TIMEOUTFACTOR for average
> developer machines.
>
> I'm OK with 35s. I'll raise it to 35s and simplify the
> comment a bit.
>
> v2
> - Raise to 35s.
>
> OK with this?

Yes.

Thanks!
diff mbox

Patch

diff --git a/nptl/tst-robust-fork.c b/nptl/tst-robust-fork.c
index d2c8a5a..c3a10da 100644
--- a/nptl/tst-robust-fork.c
+++ b/nptl/tst-robust-fork.c
@@ -25,11 +25,10 @@ 
 #include <support/xunistd.h>
 #include <sys/mman.h>
 
-/* This test can take as long as 22 seconds on some x86_64 hardware,
-   so increase the timeout to 25 seconds to allow it to pass.  More
-   than half of the time is spent in page_fault() in the kernel and
-   do_lookup_x() in ld.so.  */
-#define TIMEOUT 25
+/* This test takes as long as 20-30s on average developer hardware
+   so we increase it to 35s. More than half of the time is spent
+   in the kernel page_fault() or ld.so's do_lookup_x().  */
+#define TIMEOUT 35
 
 /* Data shared between processes. */
 struct shared