Re: droped out precise time calculations in src/interfaces/libpq/fe-connect.c - Mailing list pgsql-hackers

From Joe Conway
Subject Re: droped out precise time calculations in src/interfaces/libpq/fe-connect.c
Date
Msg-id 3DACA7CA.70509@joeconway.com
Whole thread Raw
In response to Re: droped out precise time calculations in src/interfaces/libpq/fe-connect.c  (Bruce Momjian <pgman@candle.pha.pa.us>)
Responses Re: droped out precise time calculations in src/interfaces/libpq/fe-connect.c  (Tom Lane <tgl@sss.pgh.pa.us>)
List pgsql-hackers
Tom Lane wrote:
 > Joe Conway <mail@joeconway.com> writes:
 >
 >> The thing was that with the extra +1, I was repeatedly getting a
 >> wall-clock time of 2 seconds with a timeout set to 1 second. It seemed
 >> odd to have my 1 second timeout automatically turned into 2 seconds every
 >> time.
 >
 > That is odd; seems like you should get between 1 and 2 seconds.  How were
 > you measuring the delay, exactly?

OK. I got a little more scientific about my testing. I used a php script,
running on the same machine, to connect/disconnect in a tight loop and timed
successful and unsuccessful connection attempts using microtime().

Here are the results. First with current cvs code:

current cvs libpq code
-----------------------
good connect info, using unix socket, timeout = 1 second:
=========================================================
unsuccessful 69 times: sum 0.41736388206482: avg 0.0060487519139829
successful 9931 times: sum 68.798981308937: avg 0.0069276992557584

good connect info, using hostaddr, timeout = 1 second
=====================================================
unsuccessful 72 times: sum 0.37020063400269: avg 0.0051416754722595
successful 9928 times: sum 75.047878861427: avg 0.0075592142285886

current cvs libpq code - bad hostaadr, using hostaddr, timeout = 1 second
=========================================================================
unsuccessful 100 times: sum 99.975758910179: avg 0.99975758910179
successful 0 times: sum 0: avg n/a


Clearly not good. The timeout code is causing connection failures about 0.7%
of the time. Next are the results using the attached patch. Per Bruce's
suggestion, it only adds 1 if the timeout is set to 1.


with patch libpq code
---------------------
good connect info, using unix socket, timeout = 1 second
========================================================
unsuccessful 0 times: sum 0: avg n/a
successful 10000 times: sum 68.95981669426: avg 0.006895981669426

with patch libpq code - good connect info, using hostaddr, timeout = 1 second
=============================================================================
unsuccessful 0 times: sum 0: avg n/a
successful 10000 times: sum 73.500863552094: avg 0.0073500863552094

with patch libpq code - good connect info, using hostaddr, timeout = 2 seconds
==============================================================================
unsuccessful 0 times: sum 0: avg n/a
successful 10000 times: sum 73.354710936546: avg 0.0073354710936546

with patch libpq code - bad hostaadr, using hostaddr, timeout = 1 second
========================================================================
unsuccessful 100 times: sum 149.98181843758: avg 1.4998181843758
successful 0 times: sum 0: avg n/a

with patch libpq code - bad hostaadr, using hostaddr, timeout = 2 seconds
=========================================================================
unsuccessful 100 times: sum 149.98445630074: avg 1.4998445630074
successful 0 times: sum 0: avg n/a

with patch libpq code - bad hostaadr, using hostaddr, timeout = 3 seconds
=========================================================================
unsuccessful 20 times: sum 59.842629671097: avg 2.9921314835548
successful 0 times: sum 0: avg n/a


With the patch there were 0 failures on 30000 attempts using good connect
information.

If there are no objections, please apply the attached. Otherwise let me know
if you'd like different tests or would like to try other approaches.

Thanks,

Joe


Index: src/interfaces/libpq/fe-connect.c
===================================================================
RCS file: /opt/src/cvs/pgsql-server/src/interfaces/libpq/fe-connect.c,v
retrieving revision 1.210
diff -c -r1.210 fe-connect.c
*** src/interfaces/libpq/fe-connect.c    15 Oct 2002 01:48:25 -0000    1.210
--- src/interfaces/libpq/fe-connect.c    15 Oct 2002 22:36:53 -0000
***************
*** 1066,1071 ****
--- 1066,1073 ----
      if (conn->connect_timeout != NULL)
      {
          remains.tv_sec = atoi(conn->connect_timeout);
+         if (remains.tv_sec == 1)
+             remains.tv_sec += 1;
          if (!remains.tv_sec)
          {
              conn->status = CONNECTION_BAD;

pgsql-hackers by date:

Previous
From: Bruce Momjian
Date:
Subject: Re: [GENERAL] Postgres-based system to run .org registry?
Next
From: Gavin Sherry
Date:
Subject: Vacuum improvement