Foreign key matching in PostgreSQL

Question

Just curious, if I have this table:

CREATE TABLE "post" (
    "id" SERIAL,
    "revision" INTEGER NOT NULL DEFAULT 0,
    "summary" CHARACTER VARYING NOT NULL,
    "description" TEXT NOT NULL,
    "user_id" INTEGER NOT NULL
        REFERENCES "user" ("id") MATCH FULL
            ON UPDATE CASCADE
            ON DELETE RESTRICT,
    "post_type_id" INTEGER NOT NULL
        REFERENCES "post_type" ("id") MATCH FULL
            ON UPDATE CASCADE
            ON DELETE RESTRICT,
    "ctime" TIMESTAMP WITH TIME ZONE DEFAULT NOW(),
    PRIMARY KEY("id", "revision")
);

to store posts, and this table:

CREATE TABLE "post_state" (
    "post_id" INTEGER NOT NULL,
    "assembly_seat_id" INTEGER NOT NULL
        REFERENCES "assembly_seat" ("id") MATCH FULL
            ON UPDATE CASCADE
            ON DELETE RESTRICT,
    PRIMARY KEY("post_id")
);

and I want my post_id field to point to post(id), how do I do it? I have tried with the following phrase:

    "post_id" INTEGER NOT NULL UNIQUE,
        REFERENCES "post" ("id") MATCH SIMPLE
            ON UPDATE RESTRICT
            ON DELETE RESTRICT,

but I am getting this error:

ERROR: there is no unique constraint matching given keys for referenced table "post"

The values of post_state(asembly_seat_id) do not change in this case.

You can only create a foreign key that references a single row in the other table. Since `post_id` is not unique (only the post_id/revision combination is unique), you can't reference it. — Joachim Isaksson, May 09 '13 at 12:49
@JoachimIsaksson: You may be interested in my answer that contradicts your comment in parts. — Erwin Brandstetter, May 10 '13 at 00:55
@ErwinBrandstetter yea, re-modeled the tables already :) still thanks for the answer tho — Jeffrey04, May 10 '13 at 01:09
@ErwinBrandstetter I didn't say anything about single columns in the foreign key, just that it needs to reference a _single row_ (which is exactly what you changed it to do :) ). There are (as far as I know) basically two ways of making it reference a unique row, either make it reference the composite primary key or add a unique index to the column (post.id) he was trying to reference. Adding unique breaks the primary key in post and the foreign key change breaks what I was reading that he was attempting to do, hang a post_state on a post, not a specific revision of a post. — Joachim Isaksson, May 10 '13 at 05:13
@JoachimIsaksson: Oh, you are absolutely right. I misread your "single row" as "single column". Will fix my incorrect quote immediately. My apologies. — Erwin Brandstetter, May 10 '13 at 06:41
@ErwinBrandstetter np, I realized you misread, just wanted to set things straight :) — Joachim Isaksson, May 10 '13 at 07:02

Erwin Brandstetter · Accepted Answer · 2013-05-10T06:42:14.977

A foreign key constraint can span multiple columns. You could just add the column revision to the table post_state.

CREATE TEMP TABLE post (
  post_id serial NOT NULL
 ,revision integer NOT NULL DEFAULT 0
 ,summary text NOT NULL
 ,description text NOT NULL
 ,user_id integer NOT NULL
    REFERENCES user (id) MATCH FULL ON UPDATE CASCADE ON DELETE RESTRICT
 ,post_type_id integer NOT NULL
    REFERENCES post_type (id) MATCH FULL ON UPDATE CASCADE ON DELETE RESTRICT
 ,ctime timestamptz DEFAULT NOW()
 ,PRIMARY KEY(post_id, revision)
);

CREATE TEMP TABLE post_state (
  post_id integer NOT NULL
 ,revision integer NOT NULL
 ,assembly_seat_id integer NOT NULL
    REFERENCES assembly_seat (id) MATCH FULL ON UPDATE CASCADE ON DELETE RESTRICT
 ,PRIMARY KEY(post_id, revision)
 ,FOREIGN KEY (post_id, revision) REFERENCES post (post_id, revision)
);

Read the manual about foreign key constraints.

I am using the name post_id for the primary key column of table post. Don't use id as column name. If you join a bunch of tables you end up with a bunch of columns all names id. Regrettably, some half-wit ORMs are in the habit of doing that.

Alternatively, it might be better design to have unique post_id in table post and add a table post_revision with a n:1 relation to post.

Excellent answer. SQL is easy: *once you master the syntax, everything reduces to a data modelling problem* — wildplasser, May 09 '13 at 23:15
@Erwin yea, I actually re-designed the model and have another post_revision table as you mentioned :P — Jeffrey04, May 10 '13 at 01:07

Foreign key matching in PostgreSQL

1 Answers1

Linked