PostgreSQL SQL - After Trigger assignment to NEW

Author

After Trigger assignment to NEW

Achilleus Mantzios

2006-02-25, 9:48 am

Hi,

i am currently on 7.4.12, and i what i try to do
is having an AFTER row trigger nullify one NEW column,
so as to prevent a subsequent (AFTER row) trigger from using this column's
data.

Unfortunately this nullification (assignment) does not have
any effect on the next AFTER trigger.

The first trigger is in pl/pgsql, and the second
is in C.

One possible use of the above (if it worked) would be the following.
Suppose the next AFTER triger is the (enhanced version of) dbmirror
trigger which takes care of FK constraints and navigates thru the graph
in order to mirror all depenent tables' rows too.

Suppose there is no need to mirror a specific parent table.
Then in the child table i write an AFTER trigger that nullifies
this columns which is FK to the said parent table, and so prevent
the unwanted traversal from happening in the execution of the next
AFTER trigger (dbmirror).

I'd like to ask, if it is considered the right behaviour
and if there is a plan in changing it.

Is there a reason that the NEW values should remain unchanged in AFTER
row triggers?

--
-Achilleus

---------------------------(end of broadcast)---------------------------
TIP 4: Have you searched our list archives?

http://archives.postgresql.org

Tom Lane

2006-02-25, 9:48 am

Achilleus Mantzios <achill@matrix.gatewaynet.com> writes:
> Is there a reason that the NEW values should remain unchanged in AFTER
> row triggers?

By definition, an AFTER trigger is too late to change what was stored.
Use a BEFORE trigger.

regards, tom lane

---------------------------(end of broadcast)---------------------------
TIP 9: In versions below 8.0, the planner will ignore your desire to
choose an index scan if your joining column's datatypes do not
match

Alvaro Herrera

2006-02-25, 9:48 am

Tom Lane wrote:
> Achilleus Mantzios <achill@matrix.gatewaynet.com> writes:
>
> By definition, an AFTER trigger is too late to change what was stored.
> Use a BEFORE trigger.

But a BEFORE trigger would alter the stored tuple, which is not what
Achilleus wants AFAIU.

I think the desired effect can be had by having DBMirror check the
source relation of the inserted tuple (There is a hidden attributa
called tableoid IIRC that can be used for that, I think).

--
Alvaro Herrera http://www.CommandPrompt.com/
The PostgreSQL Company - Command Prompt, Inc.

---------------------------(end of broadcast)---------------------------
TIP 2: Don't 'kill -9' the postmaster

Tom Lane

2006-02-25, 9:48 am

Alvaro Herrera < alvherre@commandprom
pt.com> writes:
> Tom Lane wrote:
[color=darkred]
> But a BEFORE trigger would alter the stored tuple, which is not what
> Achilleus wants AFAIU.

Oh, I misunderstood what he wanted ... and now that I do understand,
I think it's a really terrible idea :-(. A large part of the point
of using an AFTER trigger is to be certain you know exactly what got
stored. (BEFORE triggers can never know this with certainty because
there might be another BEFORE trigger that runs after them and edits the
tuple some more.) If one AFTER trigger could falsify the data seen by
the next, then that guarantee crumbles. For instance, a minor
programming error in a user-written trigger could break foreign-key
checking. No thanks.

> I think the desired effect can be had by having DBMirror check the
> source relation of the inserted tuple (There is a hidden attributa
> called tableoid IIRC that can be used for that, I think).

I agree --- the correct solution is to change the DBMirror triggers to
incorporate the desired filtering logic.

regards, tom lane

---------------------------(end of broadcast)---------------------------
TIP 2: Don't 'kill -9' the postmaster

Achilleus Mantzios

2006-02-25, 9:48 am

O Tom Lane Ýãñáøå óôéò Feb 24, 2006 :

> Achilleus Mantzios <achill@matrix.gatewaynet.com> writes:
>
> By definition, an AFTER trigger is too late to change what was stored.
> Use a BEFORE trigger.

Too late if someone wants to store it.
I wanna store the intented original values, thats why i use AFTER trigger.
But i would like to alter what a final AFTER trigger would see.

I'll elabarote a little.

An update happens.
The row is stored.
An after trigger is fired that alters some NEW columns
(nullifies them), aiming for a subsequent trigger
to see the altered results .

It should be something like a pointer to a HeapTuple, (right?),
so that would be feasible i suppose.

I would not even make a post if it was something that trivial.

I hope you get my point.

>
> regards, tom lane
>
> ---------------------------(end of broadcast)---------------------------
> TIP 9: In versions below 8.0, the planner will ignore your desire to
> choose an index scan if your joining column's datatypes do not
> match
>

--
-Achilleus

---------------------------(end of broadcast)---------------------------
TIP 4: Have you searched our list archives?

http://archives.postgresql.org

Achilleus Mantzios

2006-02-25, 9:48 am

O Tom Lane Ýãñáøå óôéò Feb 24, 2006 :

> Alvaro Herrera < alvherre@commandprom
pt.com> writes:
>
>
> Oh, I misunderstood what he wanted ... and now that I do understand,
> I think it's a really terrible idea :-(. A large part of the point
> of using an AFTER trigger is to be certain you know exactly what got
> stored. (BEFORE triggers can never know this with certainty because
> there might be another BEFORE trigger that runs after them and edits the
> tuple some more.) If one AFTER trigger could falsify the data seen by
> the next, then that guarantee crumbles. For instance, a minor
> programming error in a user-written trigger could break foreign-key
> checking. No thanks.

Alvaro, Tom,
thanx a lot,
i'll have to incorporate that into dbmirror.

>
>
> I agree --- the correct solution is to change the DBMirror triggers to
> incorporate the desired filtering logic.
>
> regards, tom lane
>

--
-Achilleus

---------------------------(end of broadcast)---------------------------
TIP 4: Have you searched our list archives?

http://archives.postgresql.org

Owen Jacobson

2006-02-25, 9:48 am

Achilleus Mantzios wrote:

> O Tom Lane Ýãñáøå óôéò Feb 24, 2006 :
>
>
> Too late if someone wants to store it.
> I wanna store the intented original values, thats why i use
> AFTER trigger.
> But i would like to alter what a final AFTER trigger would see.
>
> I'll elabarote a little.
>
> An update happens.
> The row is stored.
> An after trigger is fired that alters some NEW columns
> (nullifies them), aiming for a subsequent trigger
> to see the altered results .
>
> It should be something like a pointer to a HeapTuple, (right?),
> so that would be feasible i suppose.
>
> I would not even make a post if it was something that trivial.
>
> I hope you get my point.

Your real problem is that the "subsequent" trigger has behaviour you don't like. That's what you should be fixing. If dbmirror has no way to exclude specific tables from mirroring, take it up with them as a feature request, or patch dbmirror to work how
you want it to.

AFTER triggers *must* receive the row that was actually inserted/updated/deleted. If they could receive a "modified" row that didn't reflect what was actually in the database, all sorts of useful trigger-based logging and replication patterns wouldn't wo
rk, and there's really no other way to implement them. See also Tom Lane's other message for further implications of being able to modify the rows seen by AFTER triggers.

I'd also be hesitant to write triggers that have to execute in a specific order.

---------------------------(end of broadcast)---------------------------
TIP 5: don't forget to increase your free space map settings

Sponsored Links