Re: pgbench--new transaction type

On 20.06.2012 21:41, Peter Geoghegan wrote:
> On 20 June 2012 18:42, Robert Haas<robertmhaas(at)gmail(dot)com> wrote:
>> On Wed, Jun 20, 2012 at 3:48 AM, Simon Riggs<simon(at)2ndquadrant(dot)com> wrote:
>>> I'm sure Jeff submitted this because of the need for a standard test,
>>> rather than the wish to actually modify pgbench itself.
>>>
>>> Can I suggest that we include a list of standard scripts with pgbench
>>> for this purpose? These can then be copied alongside the binary when
>>> we do an install.
>>
>> I was thinking along similar lines myself. At the least, I think we
>> can't continue to add a short option for every new test type.
>> Instead, maybe we could have --test-type=WHATEVER, and perhaps that
>> then reads whatever.sql from some compiled-in directory. That would
>> allow us to sanely support a moderately large number of tests.

We could call the --test-type option -f, and the "compiled-in directory"
could be the current directory ;-).

> +1. As long as pgbench is considered to be the standard benchmarking
> tool (and I think that it is a general problem that it is), we ought
> to make an effort to give people more options.

Yeah, this sounds like a good approach. A library of standard workload
scripts seems very useful. I've been using custom scripts to benchmark
WAL insertion scalability lately, that also seems like a kind of a thing
to put in such a library. I don't know if we should ship the library of
scripts in contrib, or just put them up on a web site, but something
like that...

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com