On May 6, 2007, at 9:32 AM, Tom Lane wrote:
> Jim Nasby <decibel@decibel.org> writes:
>> There's several problems with that. First, trace_sort isn't
>> documented (or at least it's not in postgresql.conf), so most folks
>> don't know it exists. Second, in order to see it's output you have to
>> drop log_min_messages to debug. That results in a huge log volume,
>> especially on a production system.
>
> Nonsense ... trace_sort output is at LOG level.
I stand corrected. But my point still remains. It wouldn't be unusual
for a website database to be running several sorts a second; that
means 4 lines per sort, which is still a large amount of data.
If we really want to make the logfile the approved method for
monitoring performance, then why do we have the stats infrastructure
at all? It could all be replaced with logging output and pgfouine.
Why are we maintaining two separate sets of code for monitoring
database performance?
--
Jim Nasby jim@nasby.net
EnterpriseDB http://enterprisedb.com 512.569.9461 (cell)