> Hm, I think it's possible to rig the test to do dummy
> copy of pgresult, thus it's possible to see what kind
> of speed is possible.. Will try.
I added a new method (-x) to rowdump where it asks for row
with PQgetRowData() and then tries to emulate super-efficient
PGresult copy, then loads data from that PGresult.
Quick reference:
rowdump1 - single-row-mode1 [~ libpq 9.2]
rowdump2 - single-row-mode2 [~ libpq 9.1]
-s - single row mode with PQgetResult()
-z - single row mode with PQgetRowData()
-x - simulated optimized PQgetResult()
-------------------------------------------------------------
QUERY: select 10000,200,300000,rpad('x',30,'z') from
generate_series(1,5000000)
./rowdump1 -s: 6.28 6.25 6.39 avg: 6.31 [ 100.00 % ]
./rowdump2 -s: 7.49 7.40 7.57 avg: 7.49 [ 118.71 % ]
./rowdump1 -z: 2.86 2.77 2.79 avg: 2.81 [ 44.50 % ]
./rowdump1 -x: 3.46 3.27 3.29 avg: 3.34 [ 52.96 % ]
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 -s: 7.76 7.76 7.68 avg: 7.73 [ 100.00 % ]
./rowdump2 -s: 8.24 8.12 8.66 avg: 8.34 [ 107.85 % ]
./rowdump1 -z: 5.34 5.07 5.23 avg: 5.21 [ 67.41 % ]
./rowdump1 -x: 5.53 5.61 5.61 avg: 5.58 [ 72.20 % ]
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 -s: 7.49 7.66 7.59 avg: 7.58 [ 100.00 % ]
./rowdump2 -s: 7.56 8.12 7.95 avg: 7.88 [ 103.91 % ]
./rowdump1 -z: 2.77 2.76 2.76 avg: 2.76 [ 36.46 % ]
./rowdump1 -x: 3.07 3.05 3.18 avg: 3.10 [ 40.90 % ]
QUERY: select 1000,rpad('x', 400, 'z'),rpad('x', 4000, 'z') from
generate_series(1,100000)
./rowdump1 -s: 2.66 2.62 2.67 avg: 2.65 [ 100.00 % ]
./rowdump2 -s: 3.11 3.14 3.11 avg: 3.12 [ 117.74 % ]
./rowdump1 -z: 2.49 2.46 2.47 avg: 2.47 [ 93.33 % ]
./rowdump1 -x: 2.59 2.57 2.57 avg: 2.58 [ 97.23 % ]
-----------------------------------------------------------------
It shows that even if the actual "fast" row copy will be slower
than this one here, it's still quote competitive approach to
PQgetRowData(), as long it's not too slow.
So the optimized PGgetResult() may be good enough, thus we
can drop the idea of PQgetRowData().
Code attached, also in https://github.com/markokr/pqtest repo.
--
marko