Обсуждение: question on update/delete rules on views

 
I am trying to create multiple views of a single table so different groups of people have access to different subsets of records within the table.  So I need a complete set of rules for each view.  My rules for select and insert seem to work just fine, but the update/delete rules hit all the records instead of being limited by the "where" clause in my calling query.

I'm not sure if I'm doing something wrong or if I've found a bug.  Any help would be greatly appreciated.  Here is a script that demonstrates the problem:

--Create our table
drop table a;
create table a (
one char(2),
two char(2),
three int4,
primary key (one,two)
);

--Insert some data to work with
insert into a values ('aa','xz', 10);
insert into a values ('ab','xz', 12);
insert into a values ('ac','xz', 20);
insert into a values ('ad','xz', 11);
insert into a values ('ae','xz', 15);
insert into a values ('ai','xz', 30);

--Now view the data
select * from a;

--Now this view should have a valid rule for all operations
drop view view_a;
create view view_a as select one, two from a;

create rule view_a_r_insert as on insert to view_a
        do instead
                insert into a (one, two, three)
                values (new.one, new.two, 100);

--Test the insert rule
insert into view_a (one, two) values ('az','xy');

create rule view_a_r_update as on update to view_a
        do instead
                update a set two = new.two;

--Test the update rule
update view_a set two = 'mn' where one = 'az';

--Notice all records got updated--not just the one where one = 'az'
select * from a;

According to the manual (chapter 42), when you have a view with no rule qualification but and instead clause, the resultant parsetree should include the rule action, plus the qualification from the original (calling) query.  Doesn't this mean that my "where one = 'az'" clause should be appended to the update rule so that only one record gets updated?  Or am I missing something?
 

create rule view_a_r_update as on update to view_a   do instead       update a set two = new.two;

The problem is that your INSTEAD UPDATE rule is not constrained in any
way; it DOES hit every row.  Instead, do something like:
  create rule view_a_r_update as on update to view_a   do instead       update a set two = new.two       where id =
old.id;

where id is a primary key in your table.

Cheers,
Brook

Brook Milligan wrote:

>    create rule view_a_r_update as on update to view_a
>            do instead
>                    update a set two = new.two;
>
> The problem is that your INSTEAD UPDATE rule is not constrained in any
> way; it DOES hit every row.  Instead, do something like:
>
>    create rule view_a_r_update as on update to view_a
>            do instead
>                    update a set two = new.two
>                    where id = old.id;
>
> where id is a primary key in your table.
>

Thanks for the help.  The problem with your suggestion is the view has to
anticipate which column(s) the calling query wants to look at.  What if
the calling query has not specified the primary key in its where clause?
In our real case, the table has many columns.  There are a variety of
queries that act on the table based on a variety of conditions in a
variety of columns.  I'd like to avoid having to have a separate rule or
view for every possible where combination.  Maybe that is not possible,
but the manual seems to say it should work, so that's why I'm asking the
question.

>    create rule view_a_r_update as on update to view_a  >            do instead  >                    update a set
two= new.two  >                    where id = old.id;  >  > where id is a primary key in your table.
 
  Thanks for the help.  The problem with your suggestion is the view has to  anticipate which column(s) the calling
querywants to look at.  What if  the calling query has not specified the primary key in its where clause?  In our real
case,the table has many columns.  There are a variety of  queries that act on the table based on a variety of
conditionsin a  variety of columns.  I'd like to avoid having to have a separate rule or  view for every possible where
combination. Maybe that is not possible,  but the manual seems to say it should work, so that's why I'm asking the
question.

I think you misunderstand what is going on.  The original WHERE clause
(in your query) defines a set of tuples to which the UPDATE rule will
be applied.  In the example above, each of those tuples will have a
primary key value (old.id in that case) and the matching field(s) in
table (or view) a will be changed as dictated by the rule.  Thus, for
every tuple selected by your WHERE clause, the corresponding tuple in
the underlying table will be updated.  Note that as many fields as you
wish to allow updates on can be included in the set ... part of the
rule; any that are not different will just be changed to the same
value (i.e., there will be no effect).  Consequently, you don't need
lots of rules for every combination of columns (unless there are other
reasons to restrict the set of columns modifiable by different views).

Cheers,
Brook

----- Original Message -----
From: Kyle Bateman <kyle@actarg.com>
To: Brook Milligan <brook@biology.nmsu.edu>
Cc: <pgsql-sql@postgresql.org>
Sent: Wednesday, May 17, 2000 4:15 PM
Subject: Re: [SQL] question on update/delete rules on views


> Brook Milligan wrote:
>
> >    create rule view_a_r_update as on update to view_a
> >            do instead
> >                    update a set two = new.two;
> >
> > The problem is that your INSTEAD UPDATE rule is not constrained in any
> > way; it DOES hit every row.  Instead, do something like:
> >
> >    create rule view_a_r_update as on update to view_a
> >            do instead
> >                    update a set two = new.two
> >                    where id = old.id;
> >
> > where id is a primary key in your table.
> >
>
> Thanks for the help.  The problem with your suggestion is the view has to
> anticipate which column(s) the calling query wants to look at.  What if
> the calling query has not specified the primary key in its where clause?
> In our real case, the table has many columns.  There are a variety of
> queries that act on the table based on a variety of conditions in a
> variety of columns.  I'd like to avoid having to have a separate rule or
> view for every possible where combination.  Maybe that is not possible,
> but the manual seems to say it should work, so that's why I'm asking the
> question.
>
AFAIK it doesn't matter if the original query used a field - the "old" and
"new" in the rule represent the row being updated (before and after) - you
can access any column.

-- Richard Huxton

Brook Milligan wrote:

   >    create rule view_a_r_update as on update to view_a
   >            do instead
   >                    update a set two = new.two
   >                    where id = old.id;
   >
   > where id is a primary key in your table.

I think you misunderstand what is going on.  The original WHERE clause
(in your query) defines a set of tuples to which the UPDATE rule will
be applied.  In the example above, each of those tuples will have a
primary key value (old.id in that case) and the matching field(s) in
table (or view) a will be changed as dictated by the rule.  Thus, for
every tuple selected by your WHERE clause, the corresponding tuple in
the underlying table will be updated.  Note that as many fields as you
wish to allow updates on can be included in the set ... part of the
rule; any that are not different will just be changed to the same
value (i.e., there will be no effect).  Consequently, you don't need
lots of rules for every combination of columns (unless there are other
reasons to restrict the set of columns modifiable by different views).
 

Thanks Brook.  That makes sense to me now.  Here's the test file that demonstrates the proper (or at least a better) approach.  I'll post the whole thing for the benefit of others (who wish to avoid looking as stupid as me).  Even with all the examples of rules and views in the docs, I had a hard time finding an example of a view that simply gives fully functional access to a single underlying table.

BTW, do you have an equally sensible explanation of how the "where condition" that is part of the rule syntax differs from the where clause that comes after the "do instead"?

drop table a;

create table a (
one char(2),
two char(2),
three int4,
primary key (one,two)
);

insert into a values ('aa','xz', 10);
insert into a values ('ab','xz', 12);
insert into a values ('ac','xz', 100);
insert into a values ('ad','xz', 11);
insert into a values ('ae','xz', 15);
insert into a values ('ai','xz', 30);
 

drop view view_a;
create view view_a as select one, two, three from a;

create rule view_a_r_insert as on insert to view_a
    do instead
        insert into a (one, two, three)
        values (new.one, new.two, 100);
 

insert into view_a (one, two) values ('az','xy');

create rule view_a_r_update as on update to view_a
    do instead
        update a set one = new.one, two = new.two, three = new.three
        where one = old.one and two = old.two;

select * from a;
update view_a set two = 'mn' where one = 'az';
select * from a;
update view_a set two = 'mo' where three = 100;
select * from a;
 
 

BTW, do you have an equally sensible explanation of how the "where condition"  that is part of the rule syntax
differsfrom the where clause that comes after  the "do instead"?
 

I don't know what you mean by these two different where conditions.
To my understanding the only instance of a where condition in the rule
is in the "do" part.  Think of triggering the "do" part individually
for each view tuple to be updated.  If you wish to act on a table, you
must identify how the parts of the current tuple identify a tuple or
tuples in the target table.  That is the sole role of the "where"
condition in the "do" part of the rule.

Cheers,
Brook

Kyle Bateman <kyle@actarg.com> writes:
> BTW, do you have an equally sensible explanation of how the "where
> condition" that is part of the rule syntax differs from the where
> clause that comes after the "do instead"?

The rule WHERE condition determines *whether* the rule is fired for
update of a given view tuple.  (If it isn't, then that tuple is
processed as if there were no rule.)  Assuming that the rule fires,
a WHERE inside its action will determine which tuples it actually
acts on.

I think there are some examples in the docs of uses of nonempty rule
WHERE conditions.  If not, you could try looking at the rules regression
test for examples ...
        regards, tom lane