1.3.1 tarballs up for testing/signing (Again)

eg wrote:
> Branko Čibej wrote:
>
>>> Well, I can't sign the 1.3.1 Windows ZIP file.
Testing on my 
>>> 2xAthlon MP box, 5 combinations pass, but I see
random deadlocks 
>>> during the svnserve + bdb.tests. I haven't
been able to determine 
>>> what's going on, though. Both svnserve.exe and
svn.exe seem to be 
>>> idle, not waiting for anybody ...
>>>
>>> Yes, I am running an antivirus scanner on my
box, but I disabled it 
>>> for this test, and nothing changed.
>>>
>>> Most weird.
>> Heh. I may not be smart enough to figure it out,
but persistence pays 
>> off -- if the tests fail, repeat them until they
pass. And here's 
>> proof that Windows is nothing but a heisenbug. 
>>
>> Tested the .zip flle 6 ways on Windows
(httpd-2.0.54, neon-0.25.5), 
>> all tests pass.
>>
>
> Does the BDB code have any multi-threaded code?
> It sure screams out as a synchronization bug in the
software... these 
> always show up faster in multi-processor systems... and
they may not 
> be 100% repeatable in a short test sequence.
Yes, but which software? Svnserve on WIndows is
multi-threaded, the 
threads don't interact except within BDB itself. If BDB has
a 
synchronisaton problem, the best we can do is upgrade to a
version that 
doesn't.

> Isn't it too easy to ignore this error as a Windows
flakiness issue; 
> especially when others have seen it intermittently?
Since this problem seems to have been around for a while,
and nobody 
complained very much, I don't think we should hold up a
release because 
of it.

Anyway, svn-1.4.0 will have a number of BDB-related changes
in it, and 
the same tests on the same machine didn't show any
problems.

-- Brane


------------------------------------------------------------
---------
To unsubscribe, e-mail: dev-unsubscribesubversion.tigris.org
For additional commands, e-mail: dev-helpsubversion.tigris.org

Branko Čibej wrote:


> Yes, but which software? Svnserve on WIndows is
multi-threaded, the 
> threads don't interact except within BDB itself. If
BDB has a 
> synchronisaton problem, the best we can do is upgrade
to a version that 
> doesn't.

Ahhh.

> 
> Since this problem seems to have been around for a
while, and nobody 
> complained very much, I don't think we should hold up
a release because 
> of it.

I was more wondering whether a (possible) bug should be
entered.

> 
> Anyway, svn-1.4.0 will have a number of BDB-related
changes in it, and 
> the same tests on the same machine didn't show any
problems.

If I understand you correctly, the same tests were also able
to run on 
1.3.1 after a few tries. "If" there is a
"potential deadlock" issue, it 
may or may not occur every time the test is run, depending
on the nature 
of the problem, initial conditions, what else is running..

If someone is interested in trying to track it down (or see
if they 
can), it may require running tests over a longer period of
time on 
multiple cpu machines perhaps with vary load. I don't have
a multi-cpu 
or dual-core box at this time, so I can't assist.



------------------------------------------------------------
---------
To unsubscribe, e-mail: dev-unsubscribesubversion.tigris.org
For additional commands, e-mail: dev-helpsubversion.tigris.org