Re: PostgreSQL on Cygwin

Terry,

On Tue, Jul 03, 2001 at 09:08:09PM +0000, Terry Carlin wrote:
> I posted this message on the postgresql.ports.cygwin list and have not
> heard anything from anybody. Tom Lane suggested that I contact you
> directly with this problem.

I monitor the pgsql-cygwin(at)postgresql(dot)org mailing list very closely
and I did not see any of your posts. To what mailing list does
"postgresql.ports.cygwin" correspond?

> Is there anything else I can provide to assist in solving this problem.
> What I think I have found will block PostgreSQL from being used in real
> world conditions under Cygwin.

Can you reproduce this problem with any of the tests under src/test?
Otherwise, I'm willing to download the Benchmark Factory 30 day trial
to try to reproduce your environment.

> Thanks
> Terry Carlin
> VP Q/A
> Great Bridge, LLC.
> terry(at)greatbridge(dot)com
> This is a copy of the posts
> ------------------------------------------------------------------------
> ------------------------------------------------------------------------
> -----------------------------------------------------
>
> After doing more investigation on this problem, I have discovered the
> problem probably is not in PostgreSQL.
> When I get to somewhere around 50,000 transactions with two users logged
> on, the Win2K box will not allow
> anymore processes to start. Bash complains about not being able to fork()
> and I can't even start up the Win2k Notepad thru the start menu until I
> kill the postgres backends.
> I had the Task manager running and there was plenty of memory left.

My WAG is that you have found a Cygwin resource leak -- my best guess is
open handles. Since the activity of a benchmark application is by its
very nature intensive and "long lived" as compared to someone typing
away at bash, it would be much more likely for you to stumble over a
resource leak than the typical Cygwin user.

Try adding more columns (e.g, Handle Count) in Task Manager to determine
whether or not my hypothesis is correct. Another option is to use
Performance Monitor.

If you do find a resource leak, then the next step is to isolate the
part of Cygwin that is the culprit. But, ...

> I don't think that the number of connections is the main factor. I have
> logged in 10 psql concurrent sessions from other machines and have
> executed some queries manually. I have not seen any problems doing this.
>
>
> >>>>>>>>>>>>>>>>>> Original Message <<<<<<<<<<<<<<<<<<
>
> On 6/13/01, 4:05:08 PM, Terry Carlin wrote
> regarding PostgreSQL error in pq_flush():
>
>
> > I am benchmarking PostgreSQL running on CYGWIN. I am using a tool called
> > Benchmark Factory. It allows me to put
> > a good load on a database. Benchmark Factory is using ODBC to connect to
> > PostgreSQL.

Are you sure that ODBC is not the culprit? I found the following
thread:

http://www.ca.postgresql.org/mhonarc/pgsql-bugs/2001-03/msg00062.html

Maybe ODBC is still leaking other resources.

> > I am running the A3ASP benchmark from 1 to 100 users. This is a query
> > (read) only test. It measures how fast a database
> > can retrieve random records out of a table.
>
> > Somewhere around 8 to 10 users, I start getting this message in the
> > PostgreSQL log.
>
> > pq_flush: send() failed: No buffer space available; the socket cannot
> > be connected
>
> > I am running PostgreSQL under Win2k Professional.
> > I am using release Cygwin 1.3.2 and using PostgreSQL 7.2 from the Cygwin
> > distro.

I presume that you meant PostgreSQL 7.1.2 above. Also, you should try
the latest snapshot:

http://www.cygwin.com/snapshots/

Although it's a long shot, the problem may already be fixed.

> > I don't see this problem with PostgreSQL on Linux or Sun and we connect
> > 100 users during our tests.
> > Our Benchmark Factory license is only for 100 users.
>
> > My goal is to run the same set of benchmarks that Great Bridge runs to
> > test out every release of PostgreSQL that we support.
> > This can give us an idea of the performance of PostgreSQL under Cygwin.

I would be very interested in seeing the results when and if you get them.

> > Has anyone seen this before? There is some speculation here that it
> > might be a Microsoft restriction on the number of open sockets or
> > something like that.

There are MS limitations such as a small socket backlog count (on Windows
2000 Professional this value is 5). But, this would cause transient
connection refused errors -- not prevent other apps such as Notepad
from starting.

Let me know how you make out. I will attempt to help you out as much
as possible.

Thanks,
Jason

--
Jason Tishler
Director, Software Engineering Phone: 732.264.8770 x235
Dot Hill Systems Corp. Fax: 732.264.8798
82 Bethany Road, Suite 7 Email: Jason(dot)Tishler(at)dothill(dot)com
Hazlet, NJ 07730 USA WWW: http://www.dothill.com