Hi all,
I am currently playing with the nice
pgbench tool.
I would like to build a benchmark using
pgbench with customized scenarios, in order to get something quite representative
of a real workload.
I have designed a few tables, with a
simple script to populate them, and defined 3 scenarios representing typical
transactions.
But I have the following issue. Some
tables have CHAR (or TEXT or VARCHAR) columns that belong to their primary
key and I want to include into pgbench scenarios statements with conditions
on these CHAR columns, using some random values generated by pgbench.
As pgbench \set, \setrandom or \setshell
meta-commands only manage integer variables, I tried to use SQL conditions
like:
... where my_column = to_char(:var::integer,'00009FM')
...
with var previously defined by:
\setrandom var 1 :maxvar
Having previously loaded my_column with
digits strings, I get the right result. But ... this condition cannot use
any index (tested on 9.2beta2). As a result, I get looooooong index or
table scans, which is of course not acceptable in my benchmark as it is
not representative of the real data access path :-((
Does someone has a trick to manage random
char or text variable in pgbench ?
Thanks by advance for any help.
Best regards.
Philippe Beaudoin.
