Josh,
* Josh Berkus (josh@agliodbs.com) wrote:
> On the other hand, it's still true that a high STDDEV indicates a high
> variance in the response times of a particular query, whereas a low one
> indicates that most are close to the average. While precision math
> might not work if we don't have the correct distribution, for gross DBA
> checks it's still useful. That is, I can answer the question in many
> cases of: "Does this query have a high average because of outliers, or
> because it's consisently slow?" by looking at the STDDEV.
The concern is actually the reverse issue- often the question is "is
this query ever really slow?", or "when is this query really slow?" and
those questions are not answered by stddev, min, max, nor avg.
> And FWIW, for sites where we monitor pg_stat_statements, we reset daily
> or weekly. Otherwise, the stats have no meaning.
I have wondered if we (PG) should do that by default.. I agree that
often they are much more useful when reset periodically. Of course,
having actual historical information *would* be valuable, if you could
identify the time range covered..
Thanks,
Stephen