On Jun 14, 2010, at 2:22 , Greg Smith wrote: > Florian Pflug wrote: >> To be able to asses the performance characteristics of the different >> wal-related options, I patched pgbench to show the average latency of each >> individual statement. The idea is to be able to compare the latency of the >> COMMIT with the ones of the other statements. > > That's an interesting idea, particularly given that people don't really > understand where the time is going in the standard pgbench test. Your first > bit of review feedback is that this would have to be something you could > toggle on and off, there's no way most people want to pay this penalty. If > you submit a new patch with a command line option to enable this alternate > logging format and add the result to > https://commitfest.postgresql.org/action/commitfest_view?id=6 , you can put > my name down as a reviewer and I'll take a deeper look at it as part of that.
Thank for offering to review this! Here is an updated patch that adds a command-line option (-r) to enable/disable per-command latency reporting. It also uses the INSTR_TIME infrastructure that 9.0 provides, and should work correctly in multi-threaded mode. Data is collected per-thread and summarized only for reporting to avoid locking overhead. It now shows all lines for the SQL scripts together with their latencies (zero for comments), not only those containing SQL commands. I'll add this patch to the next commitfest, and put you down as a reviewer, as you suggested. best regards, Florian Pflug
pgbench_statementlatency.patch
Description: Binary data
-- Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-hackers
