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