Re: RFD: Discarded tuple count for SeqScan nodes in EXPLAIN ANALYZE

Andres Freund wrote:
> 
> When analyzing the plan of a query I often find myself questioning 
> whether an additional index may be sensible, or if it is sensible that a 
> SeqScan is used if an index is available.
> 
> The current EXPLAIN ANALYZE only shows the number of tuples matching the 
> qualifier of an SeqScan Node - for analyzing the above situation it is 
> at least equally interesting how many tuples were read and discarded.
> 
> Therefore I produced a patch which adds a 'discarded=%f' part to the 
> analyze output.
> As this is only a RFD the implementation is a bit hackish at the moment 
> - the discarded counter is increased in execScan directly instead of a 
> helper routine in instrument.c.
> Also the discarded count is displayed in other node types as well - for 
> some there might be a sensible semantic meaning to it...
> 
> Good idea - Bad idea?

Isn't the discarded count always equal to (# of rows in table - matched 
tuples)? Seems pretty redundant to me.

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