Trigger¶

CREATE TRIGGER¶

Guidelines for Trigger Definition¶

Trigger definition provides various and powerful functionalities. Before creating a trigger, you must consider the following:

Does the trigger condition expression cause unexpected results (side effect)?

You must use the SQL statements within an expectable range.
Does the trigger action change the table given as its event target?

While this type of design is not forbidden in the trigger definition, it must be carefully applied, because a trigger can be created that falls into an infinite loop. When the trigger action modifies the event target table, the same trigger can be called again. If a trigger occurs in a statement that contains a WHERE clause, there is no side effect in the table affected by the WHERE clause.
Does the trigger cause unnecessary overhead?

If the desired action can be expressed more effectively in the source, implement it directly in the source.
Is the trigger executed recursively?

If the trigger action calls a trigger and this trigger calls the previous trigger again, a recursive loop is created in the database. If a recursive loop is created, the trigger may not be executed correctly, or the current session must be forced to terminate to break the ongoing infinite loop.
Is the trigger definition unique?

A trigger defined in the same table or the one started in the same action becomes the cause of an unrecoverable error. A trigger in the same table must have a different trigger event. In addition, trigger priority must be explicitly and unambiguously defined.

Trigger Definition¶

A trigger is created by defining a trigger target, condition and action to be performed in the CREATE TRIGGER statement. A trigger is a database object that performs a defined action when a specific event occurs in the target table.

CREATE TRIGGER [schema_name.]trigger_name
[ STATUS { ACTIVE | INACTIVE } ]
[ PRIORITY key ]
<event_time> <event_type> [<event_target>]
[ IF condition ]
EXECUTE [ AFTER | DEFERRED ] action
[COMMENT 'trigger_comment'];

<event_time> ::=
    BEFORE |
    AFTER  |
    DEFERRED

<event_type> ::=
    INSERT |
    STATEMENT INSERT |
    UPDATE |
    STATEMENT UPDATE |
    DELETE |
    STATEMENT DELETE |
    ROLLBACK |
    COMMIT

<event_target> ::=
    ON [schema_name.]table_name |
    ON [schema_name.]table_name [ (column_name) ]

<condition> ::=
    expression

<action> ::=
    REJECT |
    INVALIDATE TRANSACTION |
    PRINT message_string |
    INSERT statement |
    UPDATE statement |
    DELETE statement

schema_name: Specifies the schema name of the trigger. If omitted, the schema name of the current session is used.
trigger_name: specifies the name of the trigger to be defined.
[ STATUS { ACTIVE | INACTIVE } ]: Defines the state of the trigger (if not defined, the default value is ACTIVE).
- If ACTIVE state is specified, the trigger is executed every time the corresponding event occurs.
- If INACTIVE state is specified, the trigger is not executed even when the corresponding event occurs. The state of the trigger can be modified. For details, see ALTER TRIGGER section.
[ PRIORITY key ]: specifies a trigger priority if multiple triggers are called for an event. key must be a floating point value that is not negative. If the priority is not defined, the lowest priority 0 is assigned. Triggers having the same priority are executed in a random order. The priority of triggers can be modified. For details, see ALTER TRIGGER section.
<event_time>: specifies the point of time when the conditions and actions are executed. BEFORE, AFTER or DEFERRED can be specified. For details, see the Event Time section.
<event_type>: trigger types are divided into a user trigger and a table trigger. For details, see the Trigger Event Type section.
<event_target>: An event target is used to specify the target for the trigger to be called. For details, see the Trigger Event Target section.
<condition>: specifies the trigger condition. For details, see the Trigger Condition section.
<action>: specifies the trigger action. For details, see the Trigger Action section.
trigger_comment: specifies a trigger’s comment.

Note

DBA and DBA members can create triggers in different schemas. If a user is neither DBA nor DBA member, triggers can only be created in the schema of that user.

The following example shows how to create a trigger that rejects the update if the number of medals won is smaller than 0 when an instance of the participant table is updated. As shown below, the update is rejected if you try to change the number of gold (gold) medals that Korea won in the 2004 Olympic Games to a negative number.

CREATE TRIGGER medal_trigger
BEFORE UPDATE ON participant
IF new.gold < 0 OR new.silver < 0 OR new.bronze < 0
EXECUTE REJECT;

UPDATE participant SET gold = -5 WHERE nation_code = 'KOR'
AND host_year = 2004;

ERROR: The operation has been rejected by trigger "medal_trigger".

Event Time¶

Specifies the point of time when trigger conditions and actions are executed. The types of event time are BEFORE, AFTER and DEFERRED.

BEFORE: checks the condition before the event is processed.
AFTER: checks the condition after the event is processed.
DEFERRED: checks the condition at the end of the transaction for the event. If you specify DEFERRED, you cannot use COMMIT or ROLLBACK as the event type.

Trigger Type¶

User Trigger

A trigger relevant to a specific user of the database is called a user trigger.
A user trigger has no event target and is executed only by the owner of the trigger (the user who created the trigger).
Event types that define a user trigger are COMMIT and ROLLBACK.

Table Trigger

A trigger that has a table as the event target is called a table trigger (class trigger).
A table trigger can be seen by all users who have the SELECT authorization on a target table.
Event types that define a table trigger are instance and statement events.

Trigger Event Type¶

Instance events: An event type whose unit of operation is an instance. The types of instance (record) events are as follows:
- INSERT
- UPDATE
- DELETE
Statement events: If you define a statement event as an event type, the trigger is called only once when the trigger starts even when there are multiple objects (instances) affected by the given statement (event). The types of statement events are as follows:
- STATEMENT INSERT
- STATEMENT UPDATE
- STATEMENT DELETE
Other events: COMMIT and ROLLBACK cannot be applied to individual instances.
- COMMIT
- ROLLBACK

The following example shows how to use an instance event. The example trigger is called by each instance affected by the database update. For example, if the score values of five instances in the history table are modified, the trigger is called five times.

CREATE TABLE update_logs(event_code INTEGER, score VARCHAR(10), dt DATETIME);

CREATE TRIGGER example
BEFORE UPDATE ON history(score)
EXECUTE INSERT INTO update_logs VALUES (obj.event_code, obj.score, SYSDATETIME);

If you want the trigger to be called only once, before the first instance of the score column is updated, use the STATEMENT UPDATE type as the following example.

The following example shows how to use a statement event. If you define a statement event, the trigger is called only once before the first instance gets updated even when there are multiple instances affected by the update.

CREATE TRIGGER example
BEFORE STATEMENT UPDATE ON history(score)
EXECUTE PRINT 'There was an update on history table';

Note

You must specify the event target when you define an instance or statement event as the event type.
COMMIT and ROLLBACK cannot have an event target.

Trigger Event Target¶

An event target specifies the target for the trigger to be called. The target of a trigger event can be specified as a table or column name. If a column name is specified, the trigger is called only when the specified column is affected by the event. If a column is not specified, the trigger is called when any column of the table is affected. Only UPDATE and STATEMENT UPDATE events can specify a column as the event target.

The following example shows how to specify the score column of the history table as the event target of the example trigger.

CREATE TABLE update_logs(event_code INTEGER, score VARCHAR(10), dt DATETIME);

CREATE TRIGGER example
BEFORE UPDATE ON history(score)
EXECUTE INSERT INTO update_logs VALUES (obj.event_code, obj.score, SYSDATETIME);

Combination of Event Type and Target¶

A database event calling triggers is identified by the trigger event type and event target in a trigger definition. The following table shows the trigger event type and target combinations, along with the meaning of the CUBRID database event that the trigger event represents.

Event Type	Event Target	Corresponding Database Activity
UPDATE	Table	Trigger is called when the UPDATE statement for a table is executed.
INSERT	Table	Trigger is called when the INSERT statement for a table is executed.
DELETE	Table	Trigger is called when the DELETE statement for a table is executed.
COMMIT	None	Trigger is called when database transaction is committed.
ROLLBACK	None	Trigger is called when database transaction is rolled back.

Trigger Condition¶

You can specify whether a trigger action is to be performed by defining a condition when defining the trigger.

If a trigger condition is specified, it can be written as an independent compound expression that evaluates to true or false. In this case, the expression can contain arithmetic and logical operators allowed in the WHERE clause of the SELECT statement. The trigger action is performed if the condition is true; if it is false, action is ignored.
If a trigger condition is omitted, the trigger becomes an unconditional trigger, which refers to that the trigger action is performed whenever it is called.

The following example shows how to use a correlation name in an expression within a condition. If the event type is INSERT, UPDATE or DELETE, the expression in the condition can refer to the correlation names obj, new or old to access a specific column. This example prefixes obj to the column name in the trigger condition to show that the example trigger tests the condition based on the current value of the record column.

CREATE TRIGGER example
BEFORE UPDATE ON participant
IF new.gold < 0 OR new.silver < 0 OR new.bronze < 0
EXECUTE REJECT;

The following example shows how to use the SELECT statement in an expression within a condition. The trigger in this example uses the SELECT statement that contains an aggregate function COUNT (*) to compare the value with a constant. The SELECT statement must be enclosed in parentheses and must be placed at the end of the expression.

CREATE TRIGGER example
BEFORE INSERT ON participant
IF 1000 >  (SELECT COUNT(*) FROM participant)
EXECUTE REJECT;

Note

The expression given in the trigger condition may cause side effects on the database if a method is called while the condition is performed. A trigger condition must be constructed to avoid unexpected side effects in the database.

Correlation Name¶

You can access the column values defined in the target table by using a correlation name in the trigger definition. A correlation name is the instance that is actually affected by the database operation calling the trigger. A correlation name can also be specified in a trigger condition or action.

The types of correlation names are new, old and obj. These correlation names can be used only in instance triggers that have an INSERT, UPDATE or DELETE event.

As shown in the table below, the use of correlation names is further restricted by the event time defined for the trigger condition.

BEFORE

AFTER or DERERRED

INSERT

new

obj

UPDATE

obj

new

obj

old (AFTER)

DELETE

obj

N/A

Correlation Name	Representative Attribute Value
obj	Refers to the current attribute value of an instance. This can be used to access attribute values before an instance is updated or deleted. It is also used to access attribute values after an instance has been updated or inserted.
new	Refers to the attribute value proposed by an insert or update operation. The new value can be accessed only before the instance is actually inserted or updated.
old	Refers to the attribute value that existed prior to the completion of an update operation. This value is maintained only while the trigger is being performed. Once the trigger is completed, the old values get lost.

Trigger Action¶

A trigger action describes what to be performed if the trigger condition is true or omitted. If a specific point of time (AFTER or DEFERRED) is not given in the action clause, the action is executed at once as the trigger event.

The following is a list of actions that can be used for trigger definitions.

REJECT: discards the operation that initiated the trigger and keeps the former state of the database, if the condition is not true. Once the operation is performed, REJECT is allowed only when the action time is BEFORE because the operation cannot be rejected. Therefore, you must not use REJECT if the action time is AFTER or DERERRED.
INVALIDATE TRANSACTION: allows the event operation that called the trigger, but does not allow the transaction that contains the commit to be executed. You must cancel the transaction by using the ROLLBACK statement if it is not valid. Such action is used to protect the database from having invalid data after a data-changing event happens.
PRINT: displays trigger actions on the terminal screen in text messages, and can be used during developments or tests. The results of event operations are not rejected or discarded.
INSERT: inserts one or more new instances to the table.
UPDATE: updates one or more column values in the table.
DELETE: deletes one or more instances from the table.

The following example shows how to define an action when a trigger is created. The medal_trig trigger defines REJECT in its action. REJECT can be specified only when the action time is BEFORE.

CREATE TRIGGER medal_trig
BEFORE UPDATE ON participant
IF new.gold < 0 OR new.silver < 0 OR new.bronze < 0
EXECUTE REJECT;

Note

Trigger may fall into an infinite loop when you use INSERT in an action of a trigger where an INSERT event is defined.
If a trigger where an UPDATE event is defined runs on a partitioned table, you must be careful because the defined partition can be broken or unintended malfunction may occur. To prevent such situation, CUBRID outputs an error so that the UPDATE causing changes to the running partition is not executed. Trigger may fall into an infinite loop when you use UPDATE in an action of a trigger where an UPDATE event is defined.

Trigger’s COMMENT¶

You can specify a trigger’s comment as follows.

CREATE TRIGGER trg_ab BEFORE UPDATE on abc(c) EXECUTE UPDATE cube_ab SET sumc = sumc + 1
COMMENT 'test trigger comment';

You can see a trigger’s comment by running the below statement.

SELECT name, comment FROM db_trigger;
SELECT trigger_name, comment FROM db_trig;

Or you can see a trigger’s comment with ;sc command which displays a schema in the CSQL interpreter.

$ csql -u dba demodb

csql> ;sc tbl

To change the trigger’s comment, refer to ALTER TRIGGER syntax on the below.

ALTER TRIGGER¶

In the trigger definition, STATUS and PRIORITY options can be changed by using the ALTER statement. If you need to alter other parts of the trigger (event targets or conditional expressions), you must delete and then re-create the trigger.

ALTER TRIGGER [schema_name.]trigger_name <trigger_option> ;

<trigger_option> ::=
    STATUS { ACTIVE | INACTIVE } |
    PRIORITY key

schema_name: Specifies the schema name of the trigger. If omitted, the schema name of the current session is used.
trigger_name: specifies the name of the trigger to be changed.
STATUS { ACTIVE | INACTIVE }: changes the status of the trigger.
PRIORITY key: changes the priority.

The following example shows how to create the medal_trig trigger and then change its state to INACTIVE and its priority to 0.7.

CREATE TRIGGER medal_trig
STATUS ACTIVE
BEFORE UPDATE ON participant
IF new.gold < 0 OR new.silver < 0 OR new.bronze < 0
EXECUTE REJECT;

ALTER TRIGGER medal_trig STATUS INACTIVE;
ALTER TRIGGER medal_trig PRIORITY 0.7;

Note

Only one trigger_option can be specified in a single ALTER TRIGGER statement.
To change a table trigger, you must be the trigger owner or granted the ALTER authorization on the table where the trigger belongs.
A user trigger can only be changed by its owner. For details on trigger_option, see the CREATE TRIGGER section. The key specified together with the PRIORITY option must be a non-negative floating point value.

Trigger’s COMMENT¶

You can change a trigger’s comment by running ALTER TRIGGER syntax as below.

ALTER TRIGGER [schema_name.]trigger_name [trigger_option]
[COMMENT ‘comment_string’];

schema_name: Specifies the schema name of the trigger. If omitted, the schema name of the current session is used.
trigger_name: specifies the name of the trigger to be changed.
comment_string: specifies a trigger’s comment.

If you want to change only trigger’s comment, you can omit trigger options (trigger_option).

For trigger_option, see ALTER TRIGGER on the above.

ALTER TRIGGER trg_ab COMMENT 'new trigger comment';

DROP TRIGGER¶

You can drop a trigger by using the DROP TRIGGER statement.

DROP TRIGGER [schema_name.]trigger_name ;

schema_name: Specifies the schema name of the trigger. If omitted, the schema name of the current session is used.
trigger_name: specifies the name of the trigger to be dropped.

The following example shows how to drop the medal_trig trigger.

DROP TRIGGER medal_trig;

Note

A user trigger (i.e. the trigger event is COMMIT or ROLLBACK) can be seen and dropped only by the owner.
Only one trigger can be dropped by a single DROP TRIGGER statement. A table trigger can be dropped by a user who has an ALTER authorization on the table.

RENAME TRIGGER¶

You can change a trigger name by using the TRIGGER reserved word in the RENAME statement.

RENAME TRIGGER [schema_name.]old_trigger_name {AS | TO} [schema_name.]new_trigger_name ;

schema_name: Specifies the schema name of the trigger. If omitted, the schema name of the current session is used. The schema of the current trigger and the schema of the trigger to be changed must be the same.
old_trigger_name: specifies the current name of the trigger.
new_trigger_name: specifies the name of the trigger to be modified.

RENAME TRIGGER medal_trigger AS medal_trig;

Note

A trigger name must be unique among triggers owned by the user. However, it can be the same as the table name in the database or the name of a trigger owned by another owner.
To rename a table trigger, you must be the trigger owner or granted the ALTER authorization on the table where the trigger belongs. A user trigger can only be renamed by its user.

Deferred Condition and Action¶

A deferred trigger action and condition can be executed later or canceled. These triggers include a DEFERRED time option in the event time or action clause. If the DEFERRED option is specified in the event time and the time is omitted before the action, the action is deferred automatically.

Executing Deferred Condition and Action¶

Executes the deferred condition or action of a trigger immediately.

EXECUTE DEFERRED TRIGGER <trigger_identifier> ;

<trigger_identifier> ::=
    [schema_name.]trigger_name |
    ALL TRIGGERS

schema_name: Specifies the schema name of the trigger. If omitted, the schema name of the current session is used.
trigger_name: executes the deferred action of the trigger when a trigger name is specified.
ALL TRIGGERS: executes all currently deferred actions.

Dropping Deferred Condition and Action¶

Drops the deferred condition and action of a trigger.

DROP DEFERRED TRIGGER trigger_identifier ;

<trigger_identifier> ::=
    [schema_name.]trigger_name |
    ALL TRIGGER

schema_name: Specifies the schema name of the trigger. If omitted, the schema name of the current session is used.
trigger_name : Cancels the deferred action of the trigger when a trigger name is specified.
ALL TRIGGERS : Cancels currently deferred actions.

Granting Trigger Authorization¶

Trigger authorization is not granted explicitly. Authorization on the table trigger is automatically granted to the user if the authorization is granted on the event target table described in the trigger definition. In other words, triggers that have table targets (INSERT, UPDATE, etc.) are seen by all users. User triggers (COMMIT and ROLLBACK) are seen only by the user who defined the triggers. All authorizations are automatically granted to the trigger owner.

Note

To define a table trigger, you must have an ALTER authorization on the table.
To define a user trigger, the database must be accessed by a valid user.

Trigger on REPLACE and INSERT … ON DUPLICATE KEY UPDATE¶

When the REPLACE statement and INSERT … ON DUPLICATE KEY UPDATE statements are executed, the trigger is executed in CUBRID, while DELETE, UPDATE, INSERT jobs occur internally. The following table shows the order in which the trigger is executed in CUBRID depending on the event that occurred when the REPLACE or INSERT … ON DUPLICATE KEY UPDATE statement is executed. Both the REPLACE statement and the INSERT … ON DUPLICATE KEY UPDATE statement do not execute triggers in the inherited class (table).

Execution Sequence of Triggers in the REPLACE and the INSERT … ON DUPLICATE KEY UPDATE statements

Event	Execution Sequence of Triggers
REPLACE When a record is deleted and new one is inserted	BEFORE DELETE > AFTER DELETE > BEFORE INSERT > AFTER INSERT
INSERT … ON DUPLICATE KEY UPDATE When a record is updated	BEFORE UPDATE > AFTER UPDATE
REPLACE, INSERT … ON DUPLCATE KEY UPDATE Only when a record is inserted	BEFORE INSERT > AFTER INSERT

The following example shows that INSERT … ON DUPLICATE KEY UPDATE and REPLACE are executed in the with_trigger table and records are inserted to the trigger_actions table as a consequence of the execution.

CREATE TABLE with_trigger (id INT UNIQUE);
INSERT INTO with_trigger VALUES (11);

CREATE TABLE trigger_actions (val INT);

CREATE TRIGGER trig_1 BEFORE INSERT ON with_trigger EXECUTE INSERT INTO trigger_actions VALUES (1);
CREATE TRIGGER trig_2 BEFORE UPDATE ON with_trigger EXECUTE INSERT INTO trigger_actions VALUES (2);
CREATE TRIGGER trig_3 BEFORE DELETE ON with_trigger EXECUTE INSERT INTO trigger_actions VALUES (3);

INSERT INTO with_trigger VALUES (11) ON DUPLICATE KEY UPDATE id=22;

SELECT * FROM trigger_actions;

          va
==============
            2

REPLACE INTO with_trigger VALUES (22);

SELECT * FROM trigger_actions;

          va
==============
            2
            3
            1

Trigger Debugging¶

Once a trigger is defined, it is recommended to check whether it is running as intended. Sometimes the trigger takes more time than expected in processing. This means that it is adding too much overhead to the system or has fallen into a recursive loop. This section explains several ways to debug the trigger.

The following example shows a trigger that was defined to fall into a recursive loop_tgr when it is called. A loop_tgr trigger is somewhat artificial in its purpose; it can be used as an example of debugging trigger.

CREATE TRIGGER loop_tgr
BEFORE UPDATE ON participant(gold)
IF new.gold > 0
EXECUTE UPDATE participant
        SET gold = new.gold - 1
        WHERE nation_code = obj.nation_code AND host_year = obj.host_year;

Viewing Trigger Execution Log¶

You can view the execution log of the trigger from a terminal by using the SET TRIGGER TRACE statement.

SET TRIGGER TRACE <switch> ;

<switch> ::=
    ON |
    OFF

ON: executes TRACE until the switch is set to OFF or the current database session terminates.
OFF: stops the TRACE.

The following example shows how to execute the TRACE and the loop_tgr trigger to view the trigger execution logs. To identify the trace for each condition and action executed when the trigger is called, a message is displayed on the terminal. The following message appears 15 times because the loop_tgr trigger is executed until the gold value becomes 0.

SET TRIGGER TRACE ON;
UPDATE participant SET gold = 15 WHERE nation_code = 'KOR' AND host_year = 1988;

TRACE: Evaluating condition for trigger "loop".
TRACE: Executing action for trigger "loop".

Limiting Nested Trigger¶

With the MAXIMUM DEPTH keyword of the SET TRIGGER statement, you can limit the number of triggers to be initiated at each step. By doing so, you can prevent a recursively called trigger from falling into an infinite loop.

SET TRIGGER [ MAXIMUM ] DEPTH count ;

count: A positive integer value that specifies the number of times that a trigger can recursively start another trigger or itself. If the number of triggers reaches the maximum depth, the database request stops(aborts) and the transaction is marked as invalid. The specified DEPTH applies to all other triggers except the current session. The maximum value is 32.

The following example shows how to configure the maximum number of times of recursive trigger calling to 10. This applies to all triggers that start subsequently. In this example, the gold column value is updated to 15, so the trigger is called 16 times in total. This exceeds the currently set maximum depth and the following error message occurs.

SET TRIGGER MAXIMUM DEPTH 10;
UPDATE participant SET gold = 15 WHERE nation_code = 'KOR' AND host_year = 1988;

ERROR: Maximum trigger depth 10 exceeded at trigger "loop_tgr".

Trigger Example¶

This section covers trigger definitions in the demo database. The triggers created in the demodb database are not complex, but use most of the features available in CUBRID. If you want to maintain the original state of the demodb database when testing such triggers, you must perform a rollback after changes are made to the data.

Triggers created by the user in the own database can be as powerful as applications created by the user.

The following trigger created in the participant table rejects an update to the medal column (gold, silver, bronze) if a given value is smaller than 0. The evaluation time must be BEFORE because a correlation name new is used in the trigger condition. Although not described, the action time of this trigger is also BEFORE.

CREATE TRIGGER medal_trigger
BEFORE UPDATE ON participant
IF new.gold < 0 OR new.silver < 0 OR new.bronze < 0
EXECUTE REJECT;

The trigger medal_trigger starts when the number of gold (gold) medals of the country whose nation code is ‘BLA’ is updated. Since the trigger created does not allow negative numbers, the example below will not be updated.

UPDATE participant
SET gold = -10
WHERE nation_code = 'BLA';

The following trigger has the same condition as the one above except that STATUS ACTIVE is added. If the STATUS statement is omitted, the default value is ACTIVE. You can change STATUS to INACTIVE by using the ALTER TRIGGER statement.

You can specify whether or not to execute the trigger depending on the STATUS value.

CREATE TRIGGER medal_trig
STATUS ACTIVE
BEFORE UPDATE ON participant
IF new.gold < 0 OR new.silver < 0 OR new.bronze < 0
EXECUTE REJECT;

ALTER TRIGGER medal_trig
STATUS INACTIVE;

The following trigger shows how integrity constraint is enforced when a transaction is committed. This example is different from the previous ones, in that one trigger can have specific conditions for multiple tables.

CREATE TRIGGER check_null_first
BEFORE COMMIT
IF 0 < (SELECT count(*) FROM athlete WHERE gender IS NULL)
OR 0 < (SELECT count(*) FROM game WHERE nation_code IS NULL)
EXECUTE REJECT;

The following trigger delays the update integrity constraint check for the record table until the transaction is committed. Since the DEFERRED keyword is given as the event time, the trigger is not executed at the time.

CREATE TRIGGER deferred_check_on_record
DEFERRED UPDATE ON record
IF obj.score = '100'
EXECUTE INVALIDATE TRANSACTION;

Once completed, the update in the record table can be confirmed at the last point (commit or rollback) of the current transaction. The correlation name old cannot be used in the conditional clause of the trigger where DEFERRED UPDATE is used. Therefore, you cannot create a trigger as the following.

CREATE TABLE foo (n int);
CREATE TRIGGER foo_trigger
    DEFERRED UPDATE ON foo
    IF old.n = 100
    EXECUTE PRINT 'foo_trigger';

If you try to create a trigger as shown above, an error message is displayed and the trigger fails.

ERROR: Error compiling condition for 'foo_trigger' : old.n is not defined.

The correlation name old can be used only with AFTER.