Re: ecpg: how select/insert n rows (array) in one query

On Sat, Jun 10, 2000 at 05:49:42PM +0300, Jorma O. Tähtinen wrote:
> 1) select - this is simple (well documented)
> ...

>     EXEC SQL BEGIN DECLARE SECTION;
>     struct data_t {
>         char md5[100][32];
>         int  size[100];
>     } data;
> ...

>     Only question here is: why can't I declare that data struct like
>     struct data_t {
>         char md5[32];
>         int size;
>     } data[100];
>     Is this possible and how to use it.

Hmm, it certainly should be possible. I will look into it as soon as I find
some spare time. I do not remember if there was a real problem with this.

>     if I have understod docs correctly : use of insert like select in
> about example is NOT possible
>     please somebody say that I'm wrong - small example would be nice to
> see ;-)

selects work because the backend returns more than one tuple at a time.
However, inserts do not work that way. An SQL insert command is designed to
insert one tuple. I wouldn't know which syntax to use to do that sort of
bulk loading.

>     And yes I have tried to insert multiple rows in one transaction
> block like
>     EXEC SQL BEGIN TRANSACTION
>         for(....) {
>             insert .....
>         }
>     EXEC SQL COMMIT
> 
>     but this is still way to slow at max 50-100 insert per sec. I
> understand that inserting is slow operation

Try starting the backend without -F and putting each insert into its own
transaction. Keeping lots of updates/insert in one transaction always slows
down a system. The very same would happen to mysql if it used transactions. 

>     And yes, I really want to use c+embedded sql code to do inserting
> (real app/table struct of course more complex)

If this is a ecpg problem we surely will fix it. But I doubt the performance
is better when using psql to enter the data.

Michael
-- 
Michael Meskes
Michael@Fam-Meskes.De
Go SF 49ers! Go Rhein Fire!
Use Debian GNU/Linux! Use PostgreSQL!