Re: [patch] libpq one-row-at-a-time API - Mailing list pgsql-hackers

Here is a simple test program that takes a SELECT
query, reads it and outputs a COPY-formatted stream
to standard output, to simulate some activity.

It operates on 3 modes, specified by commant-line switches:

-f   Load full resultset at once - old way.
-s   Single-Row mode using PQgetResult().
-z   Single-Row mode using PQgetRowData().

It is compiled with 2 different libpqs that correspond to
single-row-modeX branches in my github repo:

rowdump1 - libpq with rowBuf + PQgetRowData().   rowBuf is
           required for PQgetRowData.
           [ https://github.com/markokr/postgres/tree/single-row-mode1 ]

rowdump2 - Plain libpq patched for single-row mode.
           No PQgetRowData() here.
           [ https://github.com/markokr/postgres/tree/single-row-mode2 ]

Notes:

* Hardest part is picking realistic queries that matter.
  It's possible to construct artificial queries that make
  results go either way.

* It does not make sense for compare -f with others.  But it
  does make sense to compare -f from differently patched libpqs
  to detect any potential slowdowns.

* The time measured is User Time of client process.

-------------------------------------------------------
QUERY: select 10000,200,300000,rpad('x',30,'z') from
generate_series(1,5000000)
./rowdump1 -f:   3.90   3.90   3.93  avg:  3.91
./rowdump2 -f:   4.03   4.13   4.05  avg:  4.07
./rowdump1 -s:   6.26   6.33   6.49  avg:  6.36
./rowdump2 -s:   7.48   7.46   7.50  avg:  7.48
./rowdump1 -z:   2.88   2.90   2.79  avg:  2.86
QUERY: select
rpad('x',10,'z'),rpad('x',20,'z'),rpad('x',30,'z'),rpad('x',40,'z'),rpad('x',50,'z'),rpad('x',60,'z')
from generate_series(1,3000000)
./rowdump1 -f:   6.29   6.36   6.14  avg:  6.26
./rowdump2 -f:   6.79   6.69   6.72  avg:  6.73
./rowdump1 -s:   7.71   7.72   7.80  avg:  7.74
./rowdump2 -s:   8.14   8.16   8.57  avg:  8.29
./rowdump1 -z:   6.45   5.15   5.16  avg:  5.59
QUERY: select

1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98,99,100
from generate_series(1,800000)
./rowdump1 -f:   5.68   5.66   5.72  avg:  5.69
./rowdump2 -f:   5.69   5.84   5.67  avg:  5.73
./rowdump1 -s:   7.68   7.76   7.67  avg:  7.70
./rowdump2 -s:   7.57   7.54   7.62  avg:  7.58
./rowdump1 -z:   2.78   2.82   2.72  avg:  2.77
QUERY: select 1000,rpad('x', 400, 'z'),rpad('x', 4000, 'z') from
generate_series(1,100000)
./rowdump1 -f:   2.71   2.66   2.58  avg:  2.65
./rowdump2 -f:   3.11   3.14   3.16  avg:  3.14
./rowdump1 -s:   2.64   2.61   2.64  avg:  2.63
./rowdump2 -s:   3.15   3.11   3.11  avg:  3.12
./rowdump1 -z:   2.53   2.51   2.46  avg:  2.50
-------------------------------------------------------

Test code attached.  Please play with it.

By this test, both rowBuf and PQgetRowData() look good.

--
marko