Re: cache problem (v2)

> Yes, sorry, I phrased that wrong. Let me put it differently: your
> trigger runs only inside the transaction of the calling statement,
> unless that statement itself is inside a longer explicitly-called
> transaction. For example:
>
> t1 t2
>
> BEGIN
> UPDATE table_with_trigger
> SELECT something SELECT ...FROM trigger_effect
> INSERT something else
> COMMIT
>
> In this case, t2 does _not_ see the effects of the trigger in t1,
> because those effects are not visible until the COMMIT. But
>
> t1 t2
>
> UPDATE table_with_trigger
> SELECT something SELECT...FROM trigger_effect
> INSERT something else
>
> in this case, t2 _does_ see the effects, because the trigger's
> effects are COMMITted implicitly after the UPDATE statement.
>
Yes, that I understand, but it's not my case
I have :
table test ( int code, int qte);
t1
INSERT test values(1, 150)
call my trigger that SELECT WHERE code=3 (does not exist) and INSERT
INTO test (code=3, qte=150)
INSERT test values(2, 450)
call my trigger that SELECT WHERE code = 3 (exist) and UPDATE test
(code=3, qte=600)
....

Ok all work fine

Now I have a flat file :
1,150
2,450

COPY .... path_to_this_flat_file

code=3,qte=450
Why ?
Another error also is "duplicate key"
It's like that, into my trigger, the second SELECT return 0 rows
(SPI_processed = 0) and in this case the trigger try to INSERT instead
of UPDATE
I loose one week with this problem.
> If I read that right, you admit that you are inexperienced with the
> concepts and the software, and you are unable to show us all the
> relevant code or send us a precise description of what you are
> doing; but, you are convinced nevertheless that the problem is a bug
> or deficiency in PostgreSQL that nobody else seems to be having, and
> not a problem with your approach? I suggest you think again.
>
I read the documentation from postgres not all. But sufficiently to
start a test of a trigger.
but sure, I'm not a veteran :-)
I never say that it's a bug, I say that with COPY the trigger does not
work like with INSERT.
The event received by the trigger are the same no ?
And it's what I try to understand why the comportment of my trigger
change when the call
come from COPY or from a lot of INSERT it's all.
You give me a way with the COMMITTED READ.

I try to explain more : (tomorrow I try to put on our web site a full
example and data)
I work on a statics project, before this project are builded in 1990
with CIsam from informix.
In 2004 I migrate the development from CIsam to berkeleyDB.

Now to allow more possibilities for our users (like connection from OOo
base, calc and so on)
I try to migrate the model to postgres

The problem is that I receive about 4 millions of data by month.
and to speed up the major type of query called by our users I build a
sum of different items.
example :

Origins (company a, company b,....) I build a "Total Eurofer"
Market (France, Belgium, ....) I build a "Total All Markets"
Products (product a, product b) total product a+b

In real world : for the Import data : 30 Origins, 249 Markets, 1472
products, 360 periods (like 1999-01)
The sum are about 3 Origins, 28 Markets, 158 products

This is the job of my trigger : build the sum code to speed up the
standard query of our users.
for each origins
insert/update the base code
check if this code update a sum code if yes insert/update the sum code
for each market
insert/update the base code
check if this code update a sum code if yes do it
...

finally from about 4 millions of input data I output about 16 millions
of records

Sorry if I disturb you.
Regards

Guy