On 29 October 2010 21:52, Rob Richardson <Rob.Richardson@rad-con.com> wrote:
> A customer was reviewing the database that supports the application we have
> provided. One of the tables is very simple, but has over 16 million
> records. Here is the table's definition:
>
> CREATE TABLE feedback
> (
> charge integer,
> elapsed_time integer, -- number of elapsed minutes since data began
> recording
> tag_type character varying(24), -- Description of tag being recorded
> tag_value real, -- value of tag being recorded
> status smallint, -- PLC Status, recorded with Control PV only
> stack integer, -- Not used
> heating smallint DEFAULT 0, -- 1 for heating, 0 for cooling
> cooling smallint DEFAULT 0 -- not used
> )
>
> As you see, there is no primary key. There is a single index, as follows:
>
> CREATE INDEX feedback_charge_idx
> ON feedback
> USING btree
> (charge);
> In PGAdmin, the customer selected this table and clicked the grid on the
> toolbar, asking for all of the records in the table. After twenty minutes,
> a message box appeared saying that an unhandled exception had happened.
> There was no explanation of what the exception was. The database log does
> not contain any information about it. The PGAdmin display did show a number
> of records, leading me to believe that the error happened in PGAdmin rather
> than anywhere in PostGres.
>
> Can anyone explain what is happening?
Does WxWidgets/PgAdmin provide an overload of global operator new()
that follows the pre-standard C++ behaviour of returning a null ptr,
ala malloc()? C++ application frameworks that eschew exceptions often
do. This sounds like an unhandled std::bad_alloc exception.
Why don't we have some hard limit on the number of rows viewable in a
table? Would that really be so terrible?
--
Regards,
Peter Geoghegan