Re: Implementing Incremental View Maintenance

From: Paul Draper <paulddraper(at)gmail(dot)com>
To: pgsql-hackers(at)postgresql(dot)org
Cc: kgrittn(at)gmail(dot)com, thomas(dot)munro(at)gmail(dot)com
Subject: Re: Implementing Incremental View Maintenance
Date: 2019-09-15 17:52:22
Message-ID: CAApx4VQJHtiXXoPaEsr51awFE8GZr8m+HB_q5YV7R27bT4p82w@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

As I understand it, the current patch performs immediate IVM using AFTER
STATEMENT trigger transition tables.

However, multiple tables can be modified *before* AFTER STATEMENT triggers
are fired.

CREATE TABLE example1 (a int);
CREATE TABLE example2 (a int);

CREATE INCREMENTAL MATERIALIZED VIEW mv AS
SELECT example1.a, example2.a
FROM example1 JOIN example2 ON a;

WITH
insert1 AS (INSERT INTO example1 VALUES (1)),
insert2 AS (INSERT INTO example2 VALUES (1))
SELECT NULL;

Changes to example1 are visible in an AFTER STATEMENT trigger on example2,
and vice versa. Would this not result in the (1, 1) tuple being
"double-counted"?

IVM needs to either:

(1) Evaluate deltas "serially' (e.g. EACH ROW triggers)

(2) Have simultaneous access to multiple deltas:
delta_mv = example1 x delta_example2 + example2 x delta_example1 -
delta_example1 x delta_example2

This latter method is the "logged" approach that has been discussed for
deferred evaluation.

tl;dr It seems that AFTER STATEMENT triggers required a deferred-like
implementation anyway.

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Tomas Vondra 2019-09-15 20:02:51 Re: (Re)building index using itself or another index of the same table
Previous Message Kuntal Ghosh 2019-09-15 17:26:48 Re: POC: Cleaning up orphaned files using undo logs