Обсуждение: Re: Update columns in the same table in a deferred constraint trigger
På tirsdag 29. juli 2014 kl. 12:40:21, skrev Pavel Stehule <pavel.stehule@gmail.com>:
2014-07-29 12:36 GMT+02:00 Andreas Joseph Krogh <andreas@visena.com>:På tirsdag 29. juli 2014 kl. 12:27:32, skrev Pavel Stehule <pavel.stehule@gmail.com>:2014-07-29 12:21 GMT+02:00 Andreas Joseph Krogh <andreas@visena.com>:På tirsdag 29. juli 2014 kl. 12:12:17, skrev Pavel Stehule <pavel.stehule@gmail.com>:2014-07-29 12:05 GMT+02:00 Andreas Joseph Krogh <andreas@visena.com>:På tirsdag 29. juli 2014 kl. 12:01:48, skrev Pavel Stehule <pavel.stehule@gmail.com>:2014-07-29 11:59 GMT+02:00 Andreas Joseph Krogh <andreas@visena.com>:På tirsdag 29. juli 2014 kl. 11:56:17, skrev Pavel Stehule <pavel.stehule@gmail.com>:Hi2014-07-29 11:52 GMT+02:00 Andreas Joseph Krogh <andreas@visena.com>:Hi all.I have this simple schema:create table fisk(
name varchar primary key,
autofisk varchar
);I want to update the column "autofisk" on commit based the value of "name", so I created this trigger:CREATE OR REPLACE FUNCTION fisk_tf() returns TRIGGER AS $$
BEGIN
raise notice 'name %', NEW.name;
NEW.autofisk = NEW.name || CURRENT_TIMESTAMP::text;
RETURN NEW;
END;
$$ LANGUAGE plpgsql;CREATE CONSTRAINT TRIGGER fisk_t AFTER INSERT OR UPDATE ON fisk DEFERRABLE INITIALLY DEFERREDIt should be BEFORE INSERT OR UPDATE triggerHe he, yes - I know that will work, but I need the trigger to be run as a constraint-trigger, on commit, after all the data is populated in other tables and this table.It is not possible - Postgres can change data only before writingIs there a work-around, so I in the trigger can issue for example:update fisk set autofisk = NEW.name || CURRENT_TIMESTAMP::text where name = NEW.name;without it also triggering the trigger?theoretically yes - you can disable triggers via ALTER TABLE DISABLE TRIGGER
but then the code will be unmaintainable. Anything else is better than dependency in triggers. You should to think about different solution.
Sometimes triggers can be replaced by functions directly called from applications instead DML statements.I have tried this but the commit never returns, I think because it recursively triggers the trigger again for that modification.Will temporarily disabeling the trigger inside the trigger (in a transaction) work?I really afraid of this strategyI see, so it boils down to this being impossible at the moment.I really want this to be at the DML-level so any modification done also updates the "autofisk"-column.Are there any plans to make this work, that being modifying the same table in a trigger running on it where the modification (comming form statements inside the trigger-functino) like what I'm trying will not trigger the trigger?you can use a auxiliary column with information where are from a UPDATE. This information should be used for breaking recursion. But it is not a good solution. You do some too complex.
Why you need it?PavelHow would I use this auxiliary column? As I understand the WHERE-condition in the trigger-definition is not deferred, and evaled only once, or is this not what you propose? Can you make an example of how to use such an auxiliary-column?The reason I need this is that I will concat information from different tables based on information in the table the trigger is installed on. This information is to be updated in a column in the same table of type "tsvector" and used for searching later. I want the tsvector-column to be in the same table to be able to have a multicolumn index and avoid unnecessary JOIN'ing.I am thinking so correct solution for this solution is using a function instead trigger or redesign a schem
I need this function to be called whenever *any* modification (insert or update) is done on the main table, how do I accomplish that without using a trigger?
Thanks.
--
Andreas Joseph Krogh
CTO / Partner - Visena AS
Mobile: +47 909 56 963
Вложения
2014-07-29 12:49 GMT+02:00 Andreas Joseph Krogh <andreas@visena.com>:
På tirsdag 29. juli 2014 kl. 12:40:21, skrev Pavel Stehule <pavel.stehule@gmail.com>:2014-07-29 12:36 GMT+02:00 Andreas Joseph Krogh <andreas@visena.com>:På tirsdag 29. juli 2014 kl. 12:27:32, skrev Pavel Stehule <pavel.stehule@gmail.com>:2014-07-29 12:21 GMT+02:00 Andreas Joseph Krogh <andreas@visena.com>:På tirsdag 29. juli 2014 kl. 12:12:17, skrev Pavel Stehule <pavel.stehule@gmail.com>:2014-07-29 12:05 GMT+02:00 Andreas Joseph Krogh <andreas@visena.com>:På tirsdag 29. juli 2014 kl. 12:01:48, skrev Pavel Stehule <pavel.stehule@gmail.com>:2014-07-29 11:59 GMT+02:00 Andreas Joseph Krogh <andreas@visena.com>:På tirsdag 29. juli 2014 kl. 11:56:17, skrev Pavel Stehule <pavel.stehule@gmail.com>:Hi2014-07-29 11:52 GMT+02:00 Andreas Joseph Krogh <andreas@visena.com>:Hi all.I have this simple schema:create table fisk(
name varchar primary key,
autofisk varchar
);I want to update the column "autofisk" on commit based the value of "name", so I created this trigger:CREATE OR REPLACE FUNCTION fisk_tf() returns TRIGGER AS $$
BEGIN
raise notice 'name %', NEW.name;
NEW.autofisk = NEW.name || CURRENT_TIMESTAMP::text;
RETURN NEW;
END;
$$ LANGUAGE plpgsql;CREATE CONSTRAINT TRIGGER fisk_t AFTER INSERT OR UPDATE ON fisk DEFERRABLE INITIALLY DEFERREDIt should be BEFORE INSERT OR UPDATE triggerHe he, yes - I know that will work, but I need the trigger to be run as a constraint-trigger, on commit, after all the data is populated in other tables and this table.It is not possible - Postgres can change data only before writingIs there a work-around, so I in the trigger can issue for example:update fisk set autofisk = NEW.name || CURRENT_TIMESTAMP::text where name = NEW.name;without it also triggering the trigger?theoretically yes - you can disable triggers via ALTER TABLE DISABLE TRIGGER
but then the code will be unmaintainable. Anything else is better than dependency in triggers. You should to think about different solution.
Sometimes triggers can be replaced by functions directly called from applications instead DML statements.I have tried this but the commit never returns, I think because it recursively triggers the trigger again for that modification.Will temporarily disabeling the trigger inside the trigger (in a transaction) work?I really afraid of this strategyI see, so it boils down to this being impossible at the moment.I really want this to be at the DML-level so any modification done also updates the "autofisk"-column.Are there any plans to make this work, that being modifying the same table in a trigger running on it where the modification (comming form statements inside the trigger-functino) like what I'm trying will not trigger the trigger?you can use a auxiliary column with information where are from a UPDATE. This information should be used for breaking recursion. But it is not a good solution. You do some too complex.
Why you need it?PavelHow would I use this auxiliary column? As I understand the WHERE-condition in the trigger-definition is not deferred, and evaled only once, or is this not what you propose? Can you make an example of how to use such an auxiliary-column?The reason I need this is that I will concat information from different tables based on information in the table the trigger is installed on. This information is to be updated in a column in the same table of type "tsvector" and used for searching later. I want the tsvector-column to be in the same table to be able to have a multicolumn index and avoid unnecessary JOIN'ing.I am thinking so correct solution for this solution is using a function instead trigger or redesign a schemI need this function to be called whenever *any* modification (insert or update) is done on the main table, how do I accomplish that without using a trigger?
you will call function instead DML from application everywhere and you will prohibit direct DML
There are not too much possibilities :(
Pavel
Thanks.
Вложения
På tirsdag 29. juli 2014 kl. 13:22:36, skrev Pavel Stehule <pavel.stehule@gmail.com>:
2014-07-29 12:49 GMT+02:00 Andreas Joseph Krogh <andreas@visena.com>:På tirsdag 29. juli 2014 kl. 12:40:21, skrev Pavel Stehule <pavel.stehule@gmail.com>:2014-07-29 12:36 GMT+02:00 Andreas Joseph Krogh <andreas@visena.com>:På tirsdag 29. juli 2014 kl. 12:27:32, skrev Pavel Stehule <pavel.stehule@gmail.com>:2014-07-29 12:21 GMT+02:00 Andreas Joseph Krogh <andreas@visena.com>:På tirsdag 29. juli 2014 kl. 12:12:17, skrev Pavel Stehule <pavel.stehule@gmail.com>:2014-07-29 12:05 GMT+02:00 Andreas Joseph Krogh <andreas@visena.com>:På tirsdag 29. juli 2014 kl. 12:01:48, skrev Pavel Stehule <pavel.stehule@gmail.com>:2014-07-29 11:59 GMT+02:00 Andreas Joseph Krogh <andreas@visena.com>:På tirsdag 29. juli 2014 kl. 11:56:17, skrev Pavel Stehule <pavel.stehule@gmail.com>:Hi2014-07-29 11:52 GMT+02:00 Andreas Joseph Krogh <andreas@visena.com>:Hi all.I have this simple schema:create table fisk(
name varchar primary key,
autofisk varchar
);I want to update the column "autofisk" on commit based the value of "name", so I created this trigger:CREATE OR REPLACE FUNCTION fisk_tf() returns TRIGGER AS $$
BEGIN
raise notice 'name %', NEW.name;
NEW.autofisk = NEW.name || CURRENT_TIMESTAMP::text;
RETURN NEW;
END;
$$ LANGUAGE plpgsql;CREATE CONSTRAINT TRIGGER fisk_t AFTER INSERT OR UPDATE ON fisk DEFERRABLE INITIALLY DEFERREDIt should be BEFORE INSERT OR UPDATE triggerHe he, yes - I know that will work, but I need the trigger to be run as a constraint-trigger, on commit, after all the data is populated in other tables and this table.It is not possible - Postgres can change data only before writingIs there a work-around, so I in the trigger can issue for example:update fisk set autofisk = NEW.name || CURRENT_TIMESTAMP::text where name = NEW.name;without it also triggering the trigger?theoretically yes - you can disable triggers via ALTER TABLE DISABLE TRIGGER
but then the code will be unmaintainable. Anything else is better than dependency in triggers. You should to think about different solution.
Sometimes triggers can be replaced by functions directly called from applications instead DML statements.I have tried this but the commit never returns, I think because it recursively triggers the trigger again for that modification.Will temporarily disabeling the trigger inside the trigger (in a transaction) work?I really afraid of this strategyI see, so it boils down to this being impossible at the moment.I really want this to be at the DML-level so any modification done also updates the "autofisk"-column.Are there any plans to make this work, that being modifying the same table in a trigger running on it where the modification (comming form statements inside the trigger-functino) like what I'm trying will not trigger the trigger?you can use a auxiliary column with information where are from a UPDATE. This information should be used for breaking recursion. But it is not a good solution. You do some too complex.
Why you need it?PavelHow would I use this auxiliary column? As I understand the WHERE-condition in the trigger-definition is not deferred, and evaled only once, or is this not what you propose? Can you make an example of how to use such an auxiliary-column?The reason I need this is that I will concat information from different tables based on information in the table the trigger is installed on. This information is to be updated in a column in the same table of type "tsvector" and used for searching later. I want the tsvector-column to be in the same table to be able to have a multicolumn index and avoid unnecessary JOIN'ing.I am thinking so correct solution for this solution is using a function instead trigger or redesign a schemI need this function to be called whenever *any* modification (insert or update) is done on the main table, how do I accomplish that without using a trigger?you will call function instead DML from application everywhere and you will prohibit direct DML
There are not too much possibilities :(
I see, thanks.
--
Andreas Joseph Krogh
CTO / Partner - Visena AS
Mobile: +47 909 56 963