Обсуждение: Platform Testing - Cygwin
Having heard nothing on the list yet about the reported unsuccessful
parallel regression tests on Cygwin with 7.2b3, I thought I'd have a play
myself having found a spare few minutes.
System: Windows XP Professional, PIII 850MHz, 512Mb RAM, 32Gb disk
uname -a: CYGWIN_NT-5.1 PC20 1.3.3(0.46/3/2) 2001-09-12 23:54 i686 unknown
Sequential regression tests pass repeatedly.
Parallel regression tests appear to fail almost randomly. The best I got so
far was 3 failures (out of 79 tests), the worst was about 15. In particular
the horology & misc tests always seems to fail, whilst the others vary. With
the exception of the misc test, all failures appear to be due to failed
connections eg:
--- 1,3 ----
! psql: could not connect to server: Connection refused
! Is the server running on host localhost and accepting
! TCP/IP connections on port 65432?
The misc test fails with:
*** ./expected/misc.out Wed Dec 12 20:34:59 2001
--- ./results/misc.out Wed Dec 12 21:52:29 2001
***************
*** 567,573 ****
a_star
abstime_tbl
aggtest
- arrtest
b
b_star
box_tbl
--- 567,572 ----
***************
*** 633,641 ****
point_tbl
polygon_tbl
ramp
- random_tbl
real_city
- reltime_tbl
road
serialtest
serialtest_f2_seq
--- 632,638 ----
***************
*** 652,662 ****
timestamp_tbl
timestamptz_tbl
timetz_tbl
- tinterval_tbl
toyemp
varchar_tbl
xacttest
! (93 rows)
--SELECT name(equipment(hobby_construct(text 'skywalking', text 'mer')))
AS equip_name;
SELECT hobbies_by_name('basketball');
--- 649,658 ----
timestamp_tbl
timestamptz_tbl
timetz_tbl
toyemp
varchar_tbl
xacttest
! (89 rows)
--SELECT name(equipment(hobby_construct(text 'skywalking', text 'mer')))
AS equip_name;
SELECT hobbies_by_name('basketball');
Though again, this varies with each run - looking at misc.sql I assume that
this is because of the earlier failures?
I have no idea what's causing these connection failures, but if anyone else
has any ideas and would like me to try out anything please let me know -
assuming of course it's not too late for 7.2 yet...
Regards, Dave.
--
Dave Page (dpage@postgresql.org)
http://pgadmin.postgresql.org/
> Having heard nothing on the list yet about the reported unsuccessful
> parallel regression tests on Cygwin with 7.2b3, I thought I'd have a play
> myself having found a spare few minutes.
Tom Lane has speculated that some optimizations around our locking code
(which had been redone for 7.2) might be the culprit for problems in
Cygwin as it apparently was for AIX. He has since fixed the problems at
least under AIX.
Could you repeat the test with 7.2b4 (out today??)?.
- Thomas
> System: Windows XP Professional, PIII 850MHz, 512Mb RAM, 32Gb disk
> uname -a: CYGWIN_NT-5.1 PC20 1.3.3(0.46/3/2) 2001-09-12 23:54 i686 unknown
> Sequential regression tests pass repeatedly.
> Parallel regression tests appear to fail almost randomly...
Dave,
On Wed, Dec 12, 2001 at 10:18:57PM -0000, Dave Page wrote:
> Parallel regression tests appear to fail almost randomly. The best I got so
> far was 3 failures (out of 79 tests), the worst was about 15. In particular
> the horology & misc tests always seems to fail, whilst the others vary. With
> the exception of the misc test, all failures appear to be due to failed
> connections eg:
>
> --- 1,3 ----
> ! psql: could not connect to server: Connection refused
> ! Is the server running on host localhost and accepting
> ! TCP/IP connections on port 65432?
The above is a known MS Winsock limitation and is documented in FAQ_MSWIN:
2. make check can generate spurious regression test failures due to
overflowing the listen() backlog queue which causes connection
refused errors.
> System: Windows XP Professional, PIII 850MHz, 512Mb RAM, 32Gb disk
^^^^^^^^^^^^
Your system has a backlog limit of 5. Although a little dated, see the
following for details:
http://support.microsoft.com/support/kb/articles/Q127/1/44.asp
Jason
> -----Original Message----- > From: Jason Tishler [mailto:jason@tishler.net] > Sent: 13 December 2001 12:33 > To: Dave Page > Cc: 'pgsql-hackers@postgresql.org'; 'pgsql-cygwin@postgresql.org' > Subject: Re: [CYGWIN] Platform Testing - Cygwin > > > Dave, > > On Wed, Dec 12, 2001 at 10:18:57PM -0000, Dave Page wrote: > > Parallel regression tests appear to fail almost randomly. > The best I > > got so far was 3 failures (out of 79 tests), the worst was > about 15. > > In particular the horology & misc tests always seems to > fail, whilst > > the others vary. With the exception of the misc test, all failures > > appear to be due to failed connections eg: > > > > --- 1,3 ---- > > ! psql: could not connect to server: Connection refused > > ! Is the server running on host localhost and accepting > > ! TCP/IP connections on port 65432? > > The above is a known MS Winsock limitation and is documented > in FAQ_MSWIN: > > 2. make check can generate spurious regression test > failures due to > overflowing the listen() backlog queue which causes connection > refused errors. > > > System: Windows XP Professional, PIII 850MHz, 512Mb RAM, 32Gb disk > ^^^^^^^^^^^^ > > Your system has a backlog limit of 5. Although a little > dated, see the following for details: > > http://support.microsoft.com/support/kb/articles/Q127/1/44.asp > > Jason Aww nuts. I should have thought of that. I'll try again on a Win2K server. Slap on the wrist for not checking the docs - in my defence I'm recovering from a rather nasty cold, and I am following on from someone else's reported problem! Thanks Jason, Dave.
Dave, On Thu, Dec 13, 2001 at 01:30:05PM -0000, Dave Page wrote: > Slap on the wrist for not checking the docs - in my defence I'm recovering > from a rather nasty cold, and I am following on from someone else's reported > problem! No slap is necessary. I'm sorry that my terse response did not indicate my appreciation for taking the time to help out. Thanks, Jason
...
> > The above is a known MS Winsock limitation and is documented
> > in FAQ_MSWIN:
...
> Aww nuts. I should have thought of that. I'll try again on a Win2K server.
Jason and Dave, would y'all consider this a tested and supported
platform then? I'd like to correctly represent this in the ports list in
the docs for this release, but don't recall having seen a report such as
"Win+Cygwin work as well as they ever have"...
- Thomas
Thomas, I have not done any 7.2 testing myself, but Dave reports the following: On Wed, Dec 12, 2001 at 10:18:57PM -0000, Dave Page wrote: > System: Windows XP Professional, PIII 850MHz, 512Mb RAM, 32Gb disk > uname -a: CYGWIN_NT-5.1 PC20 1.3.3(0.46/3/2) 2001-09-12 23:54 i686 unknown > > Sequential regression tests pass repeatedly. and: On Thu, Dec 13, 2001 at 02:21:08PM -0000, Dave Page wrote: > 7.2b4 passes *all* tests both parallel and sequential on Windows 2000 > Server. > > On XP Pro, and by the sounds of it, any other non-server releases of > Windows, parallel tests will fail randomly due to Winsock backlog limit of 5 > on these systems (as pointed out by Jason Tishler and documented in > FAQ_MSWIN). On Thu, Dec 13, 2001 at 03:08:57PM +0000, Thomas Lockhart wrote: > Jason and Dave, would y'all consider this a tested and supported > platform then? I'd like to correctly represent this in the ports list in > the docs for this release, but don't recall having seen a report such as > "Win+Cygwin work as well as they ever have"... Since I trust Dave, I feel that the above is an accurate characterization of the situation. Jason
Got it. Thanks.
- Thomas
Jason Tishler wrote: >Dave, > >On Wed, Dec 12, 2001 at 10:18:57PM -0000, Dave Page wrote: > >>Parallel regression tests appear to fail almost randomly. The best I got so >>far was 3 failures (out of 79 tests), the worst was about 15. In particular >>the horology & misc tests always seems to fail, whilst the others vary. With >>the exception of the misc test, all failures appear to be due to failed >>connections eg: >> >>--- 1,3 ---- >>! psql: could not connect to server: Connection refused >>! Is the server running on host localhost and accepting >>! TCP/IP connections on port 65432? >> > >The above is a known MS Winsock limitation and is documented in FAQ_MSWIN: > > 2. make check can generate spurious regression test failures due to > overflowing the listen() backlog queue which causes connection > refused errors. > Could this not be "fixed" in client libs, by having a retry count/timeout. I guess that having libpq (or any other client) retry the initial connection would solve most of these short queue problems. >>System: Windows XP Professional, PIII 850MHz, 512Mb RAM, 32Gb disk >> > ^^^^^^^^^^^^ > >Your system has a backlog limit of 5. Although a little dated, see the >following for details: > > http://support.microsoft.com/support/kb/articles/Q127/1/44.asp > >Jason > >---------------------------(end of broadcast)--------------------------- >TIP 5: Have you checked our extensive FAQ? > >http://www.postgresql.org/users-lounge/docs/faq.html >
Hannu Krosing <hannu@tm.ee> writes:
> I guess that having libpq (or any other client) retry the initial
> connection would solve most of these short queue problems.
And get us accused of DOS attempts. Repeated connection attempts
after one has been rejected will be seen as unfriendly behavior by
a lot of people.
Microsoft clearly does not want people running servers on the non-server
versions of Windows, and I don't see why we should go out of our way
to circumvent that.
regards, tom lane
Tom Lane wrote: > > Hannu Krosing <hannu@tm.ee> writes: > > I guess that having libpq (or any other client) retry the initial > > connection would solve most of these short queue problems. > > And get us accused of DOS attempts. Repeated connection attempts > after one has been rejected will be seen as unfriendly behavior by > a lot of people. AFAIK sendmail keeps trying for days :) > Microsoft clearly does not want people running servers on the non-server > versions of Windows, and I don't see why we should go out of our way > to circumvent that. Ok. Just a thought. ------------- Hannu
> -----Original Message----- > From: Hannu Krosing [mailto:hannu@tm.ee] > Sent: 14 December 2001 18:31 > To: Dave Page > Cc: 'Zeugswetter Andreas SB SD'; lockhart@fourpalms.org; > pgsql-hackers@postgresql.org; pgsql-cygwin@postgresql.org > Subject: Re: [CYGWIN] Platform Testing - Cygwin > > > > Dave Page wrote: > > >Although the original test was in Windows XP on a single > processor box, > >the final tests that all passed were on Windows 2000 Server > running on > >a Dual PIII 933MHz box with 1Gb of RAM. The motherboard is > an MSI Pro > >694D. > > > Has anyone done any tests comparing PostgreSQL on Win32 and *NIX > platforms on > same/similar hardware ? > > I suspect that the initial connect could be slower on Win32 due to > reported slowness of > fork() there, but are there other issues ? I believe one of the guys at Greatbridge wroteup some benchmark results comparing Cygwin/*nix. I don't know where they can be found though but I've got a hunch Jason might have them(?). Regards, Dave.
On Sat, Dec 15, 2001 at 07:47:40PM -0000, Dave Page wrote:
> I believe one of the guys at Greatbridge wroteup some benchmark results
> comparing Cygwin/*nix. I don't know where they can be found though but I've
> got a hunch Jason might have them(?).
All that I can offer is the following:
http://archives.postgresql.org/pgsql-cygwin/2001-08/msg00029.php
and specifically:
On Wed, Aug 08, 2001 at 05:50:10PM +0000, Terry Carlin wrote:
> BTW, Up through 40 users, PostgreSQL under CYGWIN using the TPC-C
> benchmark performed very much the same as Linux PostgreSQL on the
> exact hardware.
Jason