问题
I want to do something like this in a PL/pgSQL function in Postgres 9.6:
INSERT INTO table1 (id, value) VALUES (1, 'a') ON CONFLICT DO NOTHING;
--- If the above statement didn't insert a new row
--- because id of 1 already existed,
--- then run the following statements
INSERT INTO table2 (table1_id, value) VALUES (1, 'a');
UPDATE table3 set (table1_id, time) = (1, now());
However, I don't know how to determine whether the first INSERT actually inserted a new row, or whether the the ON CONFLICT DO NOTHING was done.
I could do a SELECT at the beginning of the function to see whether a record with id of 1 exists in table1 before running all the SQL statements, but this would lead to race conditions I think.
回答1:
For a plpgsql function, use the special variable FOUND:
CREATE FUNCTION foo(int, text)
RETURNS void AS
$$
BEGIN
INSERT INTO table1 (id, value) VALUES ($1, $2) ON CONFLICT DO NOTHING;
IF NOT FOUND THEN
INSERT INTO table2 (table1_id, value) VALUES ($1, $2);
UPDATE table3 set (table1_id, time) = ($1, now())
WHERE ????; -- you surely don't want to update all rows in table3
END IF;
END
$$
Call:
SELECT foo(1, 'a');
FOUND is set to false if the INSERT does not actually insert any rows.
The manual about the ON CONFLICT Clause:
ON CONFLICT DO NOTHINGsimply avoids inserting a row as its alternative action.
The manual about Obtaining the Result Status
UPDATE,INSERT, andDELETEstatements setFOUNDtrue if at least one row is affected, false if no row is affected.
To be clear, this runs the later statements if a row in table1 does already exist, so the new row is not inserted. (Like you requested, but contrary to your question title.)
If you just want to check whether a row exists:
- PostgreSQL IF statement
Race condition?
If subsequent commands in the same transaction depend on the existing row in table1 (with a FK for instance), you'll want to lock it to defend against concurrent transactions deleting or updating it in the meantime. One way to do this: instead of DO NOTHING use DO UPDATE, but do not actually update the row. The row is still locked:
INSERT INTO table1 AS t (id, value)
VALUES ($1, $2)
ON CONFLICT (id) DO UPDATE -- specify unique column(s) or constraint / index
SET id = t.id WHERE FALSE; -- never executed, but locks the row
Obviously, if you can rule out concurrent transactions that might delete or update the same row in a conflicting manner, then the problem does not exist.
Detailed explanation:
- How to include excluded rows in RETURNING from INSERT ... ON CONFLICT
- Is SELECT or INSERT in a function prone to race conditions?
回答2:
Postgres has the returning clause and CTEs to do what you want:
WITH t1 as (
INSERT INTO table1 (id, value)
VALUES (1, 'a')
ON CONFLICT DO NOTHING
RETURNING *
),
t2 as (
INSERT INTO table2 (table1_id, value)
SELECT id, value
FROM (SELECT 1 as id, 'a' as value) t
WHERE NOT EXISTS (SELECT 1 FROM t1)
)
UPDATE table3
set (table1_id, time) = (1, now())
WHERE NOT EXISTS (SELECT 1 FROM t1);
The update looks strange because it updates all rows in table3.
回答3:
Maybe you mean something like this?
INSERT INTO table1 (id, value) VALUES (1, 'a') ON CONFLICT DO NOTHING;
--- If the above statement didn't insert a new row
--- because id of 1 already existed,
--- then run the following statements
affected_rows := SQL%ROWCOUNT;
IF affected_rows = 0 THEN
INSERT INTO table2 (table1_id, value) VALUES (1, 'a');
UPDATE table3 set (table1_id, time) = (1, now());
END IF
回答4:
The easiest and reliable way is with the especial variable FOUND, this way:
INSERT INTO table1 (id, value) values (1, ‘a’) on conflict do nothing;
IF FOUND THEN
--success
ELSE
--failure
END IF;
Here is the documentation of diagnosing a statement https://www.postgresql.org/docs/9.6/static/plpgsql-statements.html
来源:https://stackoverflow.com/questions/41430826/run-sql-statements-in-pl-pgsql-only-if-a-row-doesnt-exist