I have a table in Oracle with several constraints. When I insert a new record and not all constraints are valid, then Oracle raise only the "first" error. How to get all violations of my record?
CREATE TABLE A_TABLE_TEST (
COL_1 NUMBER NOT NULL,
COL_2 NUMBER NOT NULL,
COL_3 NUMBER NOT NULL,
COL_4 NUMBER NOT NULL
);
INSERT INTO A_TABLE_TEST values (1,null,null,2);
ORA-01400: cannot insert NULL into ("USER_4_8483C"."A_TABLE_TEST"."COL_2")
I would like to get something like this:
Column COL_2: cannot insert NULL
Column COL_3: cannot insert NULL
This would be also sufficient:
Column COL_2: not valid
Column COL_3: not valid
Of course I could write a trigger and check each column individually, but I like to prefer constraints rather than triggers, they are easier to maintain and don't require manually written code.
Any idea?
There no straightforward way to report all possible constraint violations. Because when Oracle stumble on first violation of a constraint, no further evaluation is possible, statement fails, unless that constraint is deferred one or the log errors clause has been included in the DML statement. But it should be noted that log errors clause won't be able to catch all possible constraint violations, just records first one.
As one of the possible ways is to:
create exceptions table. It can be done by executing ora_home/rdbms/admin/utlexpt.sql script. The table's structure is pretty simple;
disable all table constraints;
execute DMLs;
enable all constraints with exceptions into <<exception table name>> clause. If you executed utlexpt.sql script, the name of the table exceptions are going to be stored would be exceptions.
Test table:
create table t1(
col1 number not null,
col2 number not null,
col3 number not null,
col4 number not null
);
Try to execute an insert statement:
insert into t1(col1, col2, col3, col4)
values(1, null, 2, null);
Error report -
SQL Error: ORA-01400: cannot insert NULL into ("HR"."T1"."COL2")
Disable all table's constraints:
alter table T1 disable constraint SYS_C009951;
alter table T1 disable constraint SYS_C009950;
alter table T1 disable constraint SYS_C009953;
alter table T1 disable constraint SYS_C009952;
Try to execute the previously failed insert statement again:
insert into t1(col1, col2, col3, col4)
values(1, null, 2, null);
1 rows inserted.
commit;
Now, enable table's constraints and store exceptions, if there are any, in the exceptions table:
alter table T1 enable constraint SYS_C009951 exceptions into exceptions;
alter table T1 enable constraint SYS_C009950 exceptions into exceptions;
alter table T1 enable constraint SYS_C009953 exceptions into exceptions;
alter table T1 enable constraint SYS_C009952 exceptions into exceptions;
Check the exceptions table:
column row_id format a30;
column owner format a7;
column table_name format a10;
column constraint format a12;
select *
from exceptions
ROW_ID OWNER TABLE_NAME CONSTRAINT
------------------------------ ------- ------- ------------
AAAWmUAAJAAAF6WAAA HR T1 SYS_C009951
AAAWmUAAJAAAF6WAAA HR T1 SYS_C009953
Two constraints have been violated. To find out column names, simply refer to user_cons_columns data dictionary view:
column table_name format a10;
column column_name format a7;
column row_id format a20;
select e.table_name
, t.COLUMN_NAME
, e.ROW_ID
from user_cons_columns t
join exceptions e
on (e.constraint = t.constraint_name)
TABLE_NAME COLUMN_NAME ROW_ID
---------- ---------- --------------------
T1 COL2 AAAWmUAAJAAAF6WAAA
T1 COL4 AAAWmUAAJAAAF6WAAA
The above query gives us column names, and rowids of problematic records. Having rowids at hand, there should be no problem to find those records that cause constraint violation, fix them, and re-enable constraints once again.
Here is the script that has been used to generate alter table statements for enabling and disabling constraints:
column cons_disable format a50
column cons_enable format a72
select 'alter table ' || t.table_name || ' disable constraint '||
t.constraint_name || ';' as cons_disable
, 'alter table ' || t.table_name || ' enable constraint '||
t.constraint_name || ' exceptions into exceptions;' as cons_enable
from user_constraints t
where t.table_name = 'T1'
order by t.constraint_type
You would have to implement a before-insert trigger to loop through all the conditions that you care about.
Think about the situation from the database's perspective. When you do an insert, the database can basically do two things: complete the insert successfully or fail for some reason (typically a constraint violation).
The database wants to proceed as quickly as possibly and not do unnecessary work. Once it has found the first complaint violation, it knows that the record is not going into the database. So, the engine wisely returns an error and stops checking further constraints. There is no reason for the engine to get the full list of violations.
In the meantime I found a lean solution using deferred constraints:
CREATE TABLE A_TABLE_TEST (
COL_1 NUMBER NOT NULL DEFERRABLE INITIALLY DEFERRED,
COL_2 NUMBER NOT NULL DEFERRABLE INITIALLY DEFERRED,
COL_3 NUMBER NOT NULL DEFERRABLE INITIALLY DEFERRED,
COL_4 NUMBER NOT NULL DEFERRABLE INITIALLY DEFERRED
);
INSERT INTO A_TABLE_TEST values (1,null,null,2);
DECLARE
CHECK_CONSTRAINT_VIOLATED EXCEPTION;
PRAGMA EXCEPTION_INIT(CHECK_CONSTRAINT_VIOLATED, -2290);
REF_CONSTRAINT_VIOLATED EXCEPTION;
PRAGMA EXCEPTION_INIT(REF_CONSTRAINT_VIOLATED , -2292);
CURSOR CheckConstraints IS
SELECT TABLE_NAME, CONSTRAINT_NAME, COLUMN_NAME
FROM USER_CONSTRAINTS
JOIN USER_CONS_COLUMNS USING (TABLE_NAME, CONSTRAINT_NAME)
WHERE TABLE_NAME = 'A_TABLE_TEST'
AND DEFERRED = 'DEFERRED'
AND STATUS = 'ENABLED';
BEGIN
FOR aCon IN CheckConstraints LOOP
BEGIN
EXECUTE IMMEDIATE 'SET CONSTRAINT '||aCon.CONSTRAINT_NAME||' IMMEDIATE';
EXCEPTION
WHEN CHECK_CONSTRAINT_VIOLATED OR REF_CONSTRAINT_VIOLATED THEN
DBMS_OUTPUT.PUT_LINE('Constraint '||aCon.CONSTRAINT_NAME||' at Column '||aCon.COLUMN_NAME||' violated');
END;
END LOOP;
END;
It works with any check constraint (not only NOT NULL). Checking FOREIGN KEY Constraint should work as well.
Add/Modify/Delete of constraints does not require any further maintenance.
Related
I was wondering how can I add an identity column to existing oracle table? I am using oracle 11g. Suppose I have a table named DEGREE and I am going to add an identity column to that.
FYI table is not empty.
You can not do it in one step. Instead,
Alter the table and add the column (without primary key constraint)
ALTER TABLE DEGREE ADD (Ident NUMBER(10));
Fill the new column with data which will fulfill the primary key constraint (unique/not null), e.g. like
UPDATE DEGREE SET Ident=ROWNUM;
Alter the table and add the constraint to the column
ALTER TABLE DEGREE MODIFY (Ident PRIMARY KEY);
After that is done, you can set up a SEQUENCE and a BEFORE INSERT trigger to automatically set the id value for new records.
From Oracle 12c you would use an identity column.
For example, say your table is called demo and has 3 columns and 100 rows:
create table demo (col1, col2, col3)
as
select dbms_random.value(1,10), dbms_random.value(1,10), dbms_random.value(1,10)
from dual connect by rownum <= 100;
You could add an identity column using:
alter table demo add demo_id integer generated by default on null as identity;
update demo set demo_id = rownum;
Then reset the internal sequence to match the data and prevent manual inserts:
alter table demo modify demo_id generated always as identity start with limit value;
and define it as the primary key:
alter table demo add constraint demo_pk primary key (demo_id);
This leaves the new column at the end of the column list, which shouldn’t normally matter (except for tables with a large number of columns and row chaining issues), but it looks odd when you describe the table. However, we can at least tidy up the dictionary order using the invisible/visible hack:
SQL> desc demo
Name Null? Type
-------------------------------- -------- ----------------------
COL1 NUMBER
COL2 NUMBER
COL3 NUMBER
DEMO_ID NOT NULL NUMBER(38)
begin
for r in (
select column_name from user_tab_columns c
where c.table_name = 'DEMO'
and c.column_name <> 'DEMO_ID'
order by c.column_id
)
loop
execute immediate 'alter table demo modify '||r.column_name||' invisible';
execute immediate 'alter table demo modify '||r.column_name||' visible';
end loop;
end;
/
SQL> desc demo
Name Null? Type
-------------------------------- -------- ----------------------
DEMO_ID NOT NULL NUMBER(38)
COL1 NUMBER
COL2 NUMBER
COL3 NUMBER
One thing you can't do (as of Oracle 18.0) is alter an existing column to make it into an identity column, so you have to either go through a process like the one above but copying the existing values and finally dropping the old column, or else define a new table explicitly with the identity column in place and copy the data across in a separate step. Otherwise you'll get:
-- DEMO_ID column exists but is currently not an identity column:
alter table demo modify demo_id generated by default on null as identity start with limit value;
-- Fails with:
ORA-30673: column to be modified is not an identity column
add the column
alter table table_name add (id INTEGER);
create a sequence table_name_id_seq with start with clause, using number of rows in the table + 1 or another safe value(we don't want duplicate ids);
lock the table (no inserts)
alter table table_name lock exclusive mode;
fill the column
update table_name set id = rownum; --or another logic
add a trigger to automaticaly put the id on insert using the sequence(you can find examples on internet, for example this answer)
When you'll fire the create trigger the lock will be released. (it automatically commits).
Also, you may add unique constraint on the id column, it is best to do so.
For Oracle :
CREATE TABLE new_table AS (SELECT ROWNUM AS id, ta.* FROM old_table ta)
remember this id column is not auto incremented
I am trying to add a new column into an existing table, which has to be NOT NULL. It is a foreign key referring to ID of other table and I can't use default value.
I did it in couple of steps:
add the new column to the table with NULL constraint
run update scripts to update the new column with correct values for all rows
add foreign key constraint
change the new column to NOT NULL
The issue is that I want to follow a rule we have, that the DDL and DML scripts are separated (for couple of reasons). I can't figure out how to do this. Is there a way how to add a NOT NULL column into an existing table in a way that the DDL and DML would be separated?
Do step 1 and 3 together (if you do not then you risk adding values in step 2 that will not match the constraint).
ALTER TABLE table_name
ADD column_name NUMBER(8,0)
CONSTRAINT table_name__column_name__fk REFERENCES other_table(other_column);
You can also do step 4 at the same time if you make the NOT NULL constraint DEFERRABLE:
ALTER TABLE table_name
ADD column_name NUMBER(8,0)
CONSTRAINT table_name__column_name__nn NOT NULL DEFERRABLE
CONSTRAINT table_name__column_name__fk REFERENCES other_table(other_column);
Then update the data. For example, you can use a MERGE statement to update from a sub-query of pre-generated values:
MERGE INTO table_name dst
USING (
SELECT 1 AS id, 2 AS new_column_name FROM DUAL UNION ALL
SELECT 2, 2 FROM DUAL UNION ALL
SELECT 3, 1 FROM DUAL UNION ALL
SELECT 4, 4 FROM DUAL
) src
ON (dst.id = src.id)
WHEN MATCHED THEN
UPDATE
SET column_name = src.new_column_name;
Then:
COMMIT;
Which will enforce any DEFERRABLE NOT NULL constraints.
If you did not make the constraint DEFERRABLE
then change the column to NOT NULL:
ALTER TABLE table_name
MODIFY column_name NUMBER(8,0) NOT NULL;
db<>fiddle (not deferrable) (deferrable)
I have looked over the internet and their solutions wont fix my problem, hence I'm asking for help here to check if there's mistakes in my coding.
I wanted to create a temporary table populated by other source tables and then implement it into the fact table. I have checked if the data type and the parameter is matching or the sequence of the keys but still it's giving me the error
"ORA-02291: integrity constraint (SYSTEM.SYS_C007167) violated -
parent key not found"
Fact Table:
CREATE TABLE DW_ITEMS7364 (
DW_ID int not null,
ManID char(5),
WHID char(5),
STKID char(5),
Profit number,
CONSTRAINT DW_ID PRIMARY KEY (DW_ID),
FOREIGN KEY(ManID) REFERENCES DW_MANUFACTURER7364,
FOREIGN KEY(WHID) REFERENCES DW_WAREHOUSE7364,
FOREIGN KEY(StkID) REFERENCES DW_STOCKITEM7364);
CREATE SEQUENCE seq_items7364 START WITH 101 increment by 1;
CREATE TRIGGER trg_items7364 BEFORE INSERT OR UPDATE ON DW_ITEMS7364
FOR EACH ROW
BEGIN
SELECT seq_items7364.NEXTVAL
INTO :new.DW_ID
FROM dual;
END;
Temporary Table:
CREATE TABLE TEMP_TAB7364 AS( SELECT m.ManID, w.WHID, s.STKID, (s.SellingPrice-s.PurchasePrice) AS "Profit"
FROM MANUFACTURER7364 m LEFT OUTER JOIN STOCKITEM7364 s ON s.ManID = m.ManID
RIGHT OUTER JOIN WAREHOUSE7364 w on s.WHID = w.WHID WHERE s.SELLINGPRICE IS NOT NULL AND s.PURCHASEPRICE IS NOT NULL
);
These are my source tables:
CREATE TABLE MANUFACTURER7364(
ManID char(5),
ManName varchar (25),
CityID char(5) NOT NULL,
PRIMARY KEY(ManID),
FOREIGN KEY(CityID) REFERENCES CITY7364);
CREATE TABLE WAREHOUSE7364(
WHID char(5),
MaxNoOfPallets number,
CostPerPallet number,
SecurityLevel char(1),
FreezerFacilities varchar(10),
QuarantineFacilities varchar(10),
CityID char(5) NOT NULL,
PRIMARY KEY(WHID),
FOREIGN KEY(CityID) REFERENCES CITY7364);
CREATE TABLE STOCKITEM7364(
StkID char(5),
StkName varchar(20),
SellingPrice number,
PurchasePrice number,
ManID char(5) NOT NULL,
WHID char(5) NOT NULL,
PRIMARY KEY(StkID),
FOREIGN KEY(ManID) REFERENCES MANUFACTURER7364,
FOREIGN KEY(WHID) REFERENCES WAREHOUSE7364);
As far as I can tell, nothing of what you posted raises that error.
Additional drawback is the way you chose to create foreign key constraints. If you don't name it, Oracle assigns the name itself and it looks the way you posted it: SYSTEM.SYS_C007167.
SQL> create table test
2 (id_dept number,
3 id_emp number,
4 foreign key (id_dept) references dept (deptno),
5 foreign key (id_emp) references emp (empno));
Table created.
SQL> select constraint_name from user_constraints where table_name = 'TEST';
CONSTRAINT_NAME
------------------------------
SYS_C008172
SYS_C008173
SQL>
When one of these fails, looking at its name you have no idea what went wrong, unless you investigate a little bit more:
SQL> select column_name from user_cons_columns where constraint_name = 'SYS_C008173';
COLUMN_NAME
-----------------------
ID_EMP
SQL>
But, if you name the constraint, it is way simpler:
SQL> create table test
2 (id_dept number,
3 id_emp number,
4 constraint fk_test_dept foreign key (id_dept) references dept (deptno),
5 constraint fk_test_emp foreign key (id_emp) references emp (empno));
Table created.
SQL> select constraint_name from user_constraints where table_name = 'TEST';
CONSTRAINT_NAME
------------------------------
FK_TEST_DEPT
FK_TEST_EMP
SQL>
Another major drawback one notices is what's written in front of the dot, here: SYSTEM.SYS_C007167. Yes, that would be SYSTEM. Shortly, don't do that. Leave SYS and SYSTEM alone; they are powerful, they are special. Why would you take the risk of destroying the database if you (un)intentionally do something hazardous? Create another user, grant required privileges and work in that schema.
If I understood you correctly, once you create that temp table (TEMP_TAB7364), its contents is transferred into the DW_ITEMS7364 and - doing so - you hit the error.
If that's so, what's the purpose of the temp table? Insert directly into the target table and save resources. Will it fail? Of course it will, unless you change the query. How? I don't know - make sure that you don't insert values that don't exist in any of three tables used for enforcing referential integrity.
Though, as you already have the temp table, if it isn't too large, a (relatively) quick & dirty way of finding out which row is responsible for the error can be found with a loop, such as
begin
for cur_r in (select col1, col2, ... from temp_table) loop
begin
insert into target (col1, col2, ...)
values (cur_r.col1, cur_r.col2, ...);
exception
when others then
dbms_output.put_line(sqlerrm ||': '|| cur_r.col1 ||', '||cur_r.col2);
end;
end loop;
end;
The inner BEGIN-END block is here to make sure that the PL/SQL code won't exit at the first error, but will display them all. Then review those values and find the reason that makes your query invalid.
I would like to have a table in my Oracle database where the values of an attribute1 (values may change) can't be greater than the value (fixed) of attribute2.
Is it possible to enforce such a rule?
Is it possible to make a value impossible to change after the insert ?
Disallowing attribute1 from being larger than attribute2 can be done with a check constraint:
ALTER TABLE mytable
ADD CONSTRAINT attribute2_greater_check
CHECK (attribute2 >= attribute1)
Preventing update of attribute2 can be done with a trigger that raises an error:
CREATE OR REPLACE TRIGGER mytable_attribute2_update_tr
BEFORE UPDATE ON mytable
FOR EACH ROW
BEGIN
IF :NEW.attribute2 != :OLD.attribute2
THEN
RAISE_APPLICATION_ERROR(-20101, 'attribute2 cannot be updated');
END IF;
END;
/
One way to do is to use an appropriate CONSTRAINT when creating the table:
CREATE TABLE table_name
(
column1 integer,
column2 integer not null,
CONSTRAINT constraint_name CHECK (column1 <= column2)
);
INSERT INTO table_name VALUES(1,1); //ok
INSERT INTO table_name VALUES(2,1); //gives an error
Such constraints can use any fields in the table at hand, but may not access other tables via a subselect.
EDIT: I just realized you also asked another question ... that has been answered already by #Mureinik.
Does Postgres have any way to say ALTER TABLE foo ADD CONSTRAINT bar ... which will just ignore the command if the constraint already exists, so that it doesn't raise an error?
A possible solution is to simply use DROP IF EXISTS before creating the new constraint.
ALTER TABLE foo DROP CONSTRAINT IF EXISTS bar;
ALTER TABLE foo ADD CONSTRAINT bar ...;
Seems easier than trying to query information_schema or catalogs, but might be slow on huge tables since it always recreates the constraint.
Edit 2015-07-13:
Kev pointed out in his answer that my solution creates a short window when the constraint doesn't exist and is not being enforced. While this is true, you can avoid such a window quite easily by wrapping both statements in a transaction.
This might help, although it may be a bit of a dirty hack:
create or replace function create_constraint_if_not_exists (
t_name text, c_name text, constraint_sql text
)
returns void AS
$$
begin
-- Look for our constraint
if not exists (select constraint_name
from information_schema.constraint_column_usage
where table_name = t_name and constraint_name = c_name) then
execute constraint_sql;
end if;
end;
$$ language 'plpgsql'
Then call with:
SELECT create_constraint_if_not_exists(
'foo',
'bar',
'ALTER TABLE foo ADD CONSTRAINT bar CHECK (foobies < 100);')
Updated:
As per Webmut's answer below suggesting:
ALTER TABLE foo DROP CONSTRAINT IF EXISTS bar;
ALTER TABLE foo ADD CONSTRAINT bar ...;
That's probably fine in your development database, or where you know you can shut out the apps that depend on this database for a maintenance window.
But if this is a lively mission critical 24x7 production environment you don't really want to be dropping constraints willy nilly like this. Even for a few milliseconds there's a short window where you're no longer enforcing your constraint which may allow errant values to slip through. That may have unintended consequences leading to considerable business costs at some point down the road.
You can use an exception handler inside an anonymous DO block to catch the duplicate object error.
DO $$
BEGIN
BEGIN
ALTER TABLE foo ADD CONSTRAINT bar ... ;
EXCEPTION
WHEN duplicate_table THEN -- postgres raises duplicate_table at surprising times. Ex.: for UNIQUE constraints.
WHEN duplicate_object THEN
RAISE NOTICE 'Table constraint foo.bar already exists';
END;
END $$;
http://www.postgresql.org/docs/9.4/static/sql-do.html http://www.postgresql.org/docs/9.4/static/plpgsql-control-structures.html
http://www.postgresql.org/docs/9.4/static/errcodes-appendix.html
you can run query over pg_constraint table to find constraint exists or not.like:
SELECT 1 FROM pg_constraint WHERE conname = 'constraint_name'"
Creating constraints can be an expensive operation on a table containing lots of data so I recommend not dropping constraints only to immediately create them again immediately after - you only want to create that thing once.
I chose to solve this using an anonymous code block, very similar to Mike Stankavich, however unlike Mike (who catches an error) I first check to see if the constraint exists:
DO $$
BEGIN
IF NOT EXISTS ( SELECT constraint_schema
, constraint_name
FROM information_schema.check_constraints
WHERE constraint_schema = 'myschema'
AND constraint_name = 'myconstraintname'
)
THEN
ALTER TABLE myschema.mytable ADD CONSTRAINT myconstraintname CHECK (column <= 100);
END IF;
END$$;
Using information_schema.constraint_column_usage to check for the constraint doesn't work for foreign keys. I use pg_constraint to check for primary keys, foreign keys or unique constraints:
CREATE OR REPLACE FUNCTION add_constraint(t_name text, c_name text, constraint_sql text)
RETURNS void
AS $$
BEGIN
IF NOT EXISTS(
SELECT c.conname
FROM pg_constraint AS c
INNER JOIN pg_class AS t ON c.conrelid = t."oid"
WHERE t.relname = t_name AND c.conname = c_name
) THEN
EXECUTE 'ALTER TABLE ' || t_name || ' ADD CONSTRAINT ' || c_name || ' ' || constraint_sql;
END IF;
END;
$$
LANGUAGE plpgsql;
Examples:
SELECT add_constraint('client_grant_system_scopes', 'client_grant_system_scopes_pk', 'PRIMARY KEY (client_grants_id, tenant, "scope");');
SELECT add_constraint('client_grant_system_scopes', 'client_grant_system_scopes_fk', 'FOREIGN KEY (tenant,"scope") REFERENCES system_scope(tenant,"scope") ON DELETE CASCADE;');
SELECT add_constraint('jwt_assertion_issuers', 'jwt_assertion_issuers_issuer_key', 'UNIQUE (issuer);');
Take advantage of regclass to reduce verbosity, increase performance, and avoid errors related to table naming clashes between schemas:
DO $$ BEGIN
IF NOT EXISTS (SELECT FROM pg_constraint
WHERE conrelid = 'foo'::regclass AND conname = 'bar') THEN
ALTER TABLE foo ADD CONSTRAINT bar...;
END IF;
END $$;
This will also work for tables in other schemas, e.g.:
DO $$ BEGIN
IF NOT EXISTS (SELECT FROM pg_constraint
WHERE conrelid = 's.foo'::regclass AND conname = 'bar') THEN
ALTER TABLE s.foo ADD CONSTRAINT bar...;
END IF;
END $$;
In psql You can use metacommand \gexec for run generated query.
SELECT 'ALTER TABLE xx ADD CONSTRAINT abc' WHERE not EXISTS (SELECT True FROM pg_constraint WHERE conname = 'abc') \gexec
For me those solutions didn't work because the constraint was a primary key.
This one worked for me:
ALTER TABLE <table.name> DROP CONSTRAINT IF EXISTS <constraint.name> CASCADE;
Considering all the above mentioned answers , the below approach help if you just want to check if a constraint exist in the table in which you are trying to insert and raise a notice if there happens to be one
DO
$$ BEGIN
IF NOT EXISTS (select constraint_name
from information_schema.table_constraints
where table_schema='schame_name' and upper(table_name) =
upper('table_name') and upper(constraint_name) = upper('constraint_name'))
THEN
ALTER TABLE TABLE_NAME ADD CONSTRAINT CONTRAINT_NAME..... ;
ELSE raise NOTICE 'Constraint CONTRAINT_NAME already exists in Table TABLE_NAME';
END IF;
END
$$;
Don't know why so many lines of code ?
-- SELECT "Column1", "Column2", "Column3" , count(star) FROM dbo."MyTable" GROUP BY "Column1" , "Column2" , "Column3" HAVING count(*) > 1;
alter table dbo."MyTable" drop constraint if exists "MyConstraint_Name" ;
ALTER TABLE dbo."MyTable" ADD CONSTRAINT "MyConstraint_Name" UNIQUE("Column1", "Column3", "Column2");