Writeable ct es_pgcon_may_2011

Writeable CTEs,
A Revolution in SQL
PGCon
May 20, 2011
Copyright © 2011
David Fetter dfetter@vmware.com
All Rights Reserved

Current CTEs
WITH [RECURSIVE] t1 [(column type,…)] AS
(
[SELECT | VALUES]
[UNION [ALL]
[SELECT]
),
t2 AS…tn AS…
SELECT…

Travelling Salesman Problem
Given a number of cities and the costs of travelling
from any city to any other city, what is the least-
cost round-trip route that visits each city exactly
once and then returns to the starting city?

OBTW

With CTE and Windowing, SQL is Turing Complete.

What Didn't the
Old Syntax Do?

For 9.1:
Simple Partition Management

CREATE TABLE log (
ts TIMESTAMPTZ NOT NULL,
msg TEXT NOT NULL
);

For 9.1:
CREATE TABLE log_201101 ()
INHERITS(log);

ALTER TABLE log_201101 ADD
CONSTRAINT right_month CHECK(
ts >= '2011-01-01' AND
ts < '2011-02-01');

For 9.1:

berliner@pgday_eu_2010:54321# WITH t1 AS (
DELETE FROM ONLY log
WHERE ts >='2011-01-01' AND ts < '2011-02-01'
RETURNING *
)
INSERT INTO log_201101 SELECT * FROM t1;
INSERT 0 45270

Query Clustering:
I/O Minimization

CREATE TABLE person (
id SERIAL PRIMARY KEY,
first_name TEXT,
last_name TEXT,
CHECK (CASE WHEN first_name IS NULL THEN 0 ELSE 1 END +
CASE WHEN last_name IS NULL THEN 0 ELSE 1 END >= 1),
birthdate DATE NOT NULL,
gender TEXT
);

Query Clustering:
I/O Minimization

CREATE TABLE im (
provider TEXT NOT NULL, /* should be fk */
handle TEXT NOT NULL
);

Query Clustering:
I/O Minimization
CREATE TABLE phone (
country_code TEXT NOT NULL,
phone_number TEXT NOT NULL,
extension TEXT
);

Query Clustering:
I/O Minimization
CREATE TABLE street (
street1 TEXT NOT NULL,
street2 TEXT,
street3 TEXT,
city TEXT NOT NULL,
state TEXT,
country TEXT NOT NULL,
post_code TEXT
);

Query Clustering:
I/O Minimization
CREATE TABLE person_im (
person_id INTEGER NOT NULL REFERENCES person (id),
im_id INTEGER NOT NULL REFERENCES im (id),
UNIQUE (person_id, im_id)
);

CREATE TABLE person_phone (
phone_id INTEGER NOT NULL REFERENCES phone (id),
UNIQUE (person_id, phone_id)
);

CREATE TABLE person_street (
street_id INTEGER NOT NULL REFERENCES street (id),
UNIQUE (person_id, street_id)
);

Query Clustering:
I/O Minimization

WITH t_person AS (
INSERT INTO person (first_name, last_name)
VALUES ('David', 'Fetter')
RETURNING id
),

Query Clustering:
I/O Minimization
t_im AS (
INSERT INTO im (provider, handle)
VALUES
('Yahoo!', 'dfetter'),
('AIM', 'dfetter666'),
('XMPP', 'davidfetter@postgresql.org')
RETURNING id
),
t_person_im AS (
INSERT INTO person_im
SELECT * FROM t_person CROSS JOIN t_im
),

Query Clustering:
I/O Minimization
t_phone (phone_id) AS (
INSERT INTO phone (country_code, phone_number)
VALUES
('+1','415 235 3778'),
('+1','650 427 3751')
RETURNING id
),
t_person_phone AS (
INSERT INTO person_phone
SELECT * FROM t_person CROSS JOIN t_phone
),

Query Clustering:
I/O Minimization

t_street AS (
INSERT INTO street (street1, city, state, country, post_code)
VALUES
('438 W. Grand Ave., Apartment 727','Oakland','California','USA','94612-2341'),
('3421 Hillview Avenue', 'Palo Alto','California','USA','94304')
),
t_person_street AS (
INSERT INTO person_street
SELECT * FROM t_person CROSS JOIN t_street
)

Query Clustering:
I/O Minimization

VALUES(true);

Query Clustering:
Transaction Management
CREATE TABLE foo (
bar_id INTEGER NOT NULL
);

CREATE TABLE bar (
foo_id INTEGER NOT NULL REFERENCES foo(id)
ON DELETE CASCADE
INITIALLY DEFERRED
);

ALTER TABLE foo ADD FOREIGN KEY (bar_id) REFERENCES bar(id)
ON DELETE CASCADE
INITIALLY DEFERRED;

Query Clustering:
Transaction Management
WITH t AS
(
! INSERT INTO foo(id, bar_id)
! VALUES(
! ! DEFAULT,
! ! nextval(pg_get_serial_sequence('bar', 'id'))
)
! RETURNING id AS foo_id, bar_id
)
INSERT INTO bar(id,foo_id)
SELECT bar_id, foo_id FROM t RETURNING *;

Materialized Views
WITH t AS (
DELETE FROM foo
RETURNING a, b
)
INSERT INTO mv_foo
SELECT a, SUM(b)
FROM t GROUP BY a

Just Cool Stuff™
WITH t AS (
INSERT INTO foo
VALUES (...),...,(...)
RETURNING a
)
SELECT bar.a, bar.b
FROM bar JOIN t USING(a)

Head-Scratcher:
SNAPSHOTS
WITH t AS (DELETE FROM foo)
SELECT * FROM foo;

(0 rows)?
(100123 rows)?

How'd He Do That?!?

First try: David digs into the grammar and gets cut
a few times.

How'd He Do That?!?

First try: Marko reworks the planner. It needs to
know when it creates a ModifyTable node. These used
to have another name.

How'd He Do That?!?

First try: Marko reworks the executor. It needs new
nodes. Mmmm...nodes.

How'd He Do That?!?

Marko reworks the executor, Part II:
Copy & Paste. Now it's getting ugly...

How'd He Do That?!?

Jaime Casanova, Tom Lane, and Robert Haas look at
the reworked executor.

D'oh!

How'd He Do That?!?

FAIL!
Way too much code copying from top
level to the new nodes.

How'd He Do That?!?

Planner changes for ModifyTable node (a few)

How'd He Do That?!?

Executor changes: ONE new node called ModifyTable

How'd He Do That?!?

Marko Tiikkaja restructures the whole code base for the
ModifyTable node. "The usual stuff," (he said casually) for
new nodes.

Next Steps
INSERT, UPDATE and DELETE on the top level.
Finish the patch
EXPLAIN ANALYZE
SPI
SQL functions
Bug ﬁxes, cleanup, testing
RECURSIVE
Optimization

Future?

DCL (GRANT/REVOKE)
DDL (CREATE/ALTER/DROP)

Where Is It?!?

http://www.postgresql.org/developer/beta

WITH t AS (
VALUES
('Questions'),
('Comments'),
('Rotten Tomatoes')
)
INSERT INTO davidfetter
SELECT * FROM t
RETURNING...

Writeable ct es_pgcon_may_2011

More Related Content

What's hot (19)

Similar to Writeable ct es_pgcon_may_2011 (20)

More from David Fetter (16)

Recently uploaded (20)

Writeable ct es_pgcon_may_2011

Editor's Notes