[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Intermittent parallel test failures
From: |
Ralf Wildenhues |
Subject: |
Re: Intermittent parallel test failures |
Date: |
Sun, 16 Nov 2008 18:23:43 +0100 |
User-agent: |
Mutt/1.5.18 (2008-05-17) |
Hi Bob,
* Bob Proulx wrote on Sat, Nov 15, 2008 at 04:36:01AM CET:
> Ralf Wildenhues wrote:
> > Thanks for the report. This is a different type of failure than the
> > last one, I'm glad. It is of the kind of the lost race inherent and
> > unavoidable in the test: trying to find out whether parallel execution
> > really is faster than sequential, is not deterministically possible
> > on non-real-time systems.
>
> And will have quite different results on single cpu versus multi cpu
> systems.
Not as much as it may seem: the tests all consist of 'sleep 1' calls.
As long as the system is fast enough that the Autotest overhead per test
group is small compared to one second, a single cpu system should have
no problem with it.
> > Was the system very loaded at the time?
>
> I do not know and do not keep records to be able to correlate it.
[...]
> For this type of test perhaps it could be reasonable to gather this
> information dynamically? (e.g. with uptime) Useful debug info...
Good idea.
> > Note I'm not suggesting to make it less so; au contraire, this test
> > part should either be disabled or made even less prone to failure in
> > that case. This is the first such failure in surely a lot of test
> > runs.
>
> If it is a different test failure then it certainly has had a lot of
> test runs without producing this failure.
Yes. And I have never been able to provoke it myself, even on a very
loaded system like 'make check TESTSUITEFLAGS=-j'.
Cheers,
Ralf