SQL CHECK constraint to prevent date overlap

Question

I have a table that describes which software versions were installed on a machine at various times:

machine_id::integer, version::text, datefrom::timestamp, dateto::timestamp

I'd like to do a constraint to ensure that no date ranges overlap, i.e. it is not possible to have multiple software versions installed on a machine at the same time.

How can this be achieved in SQL? I am using PostgreSQL v8.4.

Answer 1

In PostgreSQL 8.4 this can only be solved with triggers. The trigger will have to check on insert/update that no conflicting rows exist. Because transaction serializability doesn't implement predicate locking you'll have to do the necessary locking by yourself. To do that SELECT FOR UPDATE the row in the machines table so that no other transaction could be concurrently inserting data that might conflict.

In PostgreSQL 9.0 there will be a better solution to this, called exclusion constraints (somewhat documented under CREATE TABLE). That will let you specify a constraint that date ranges must not overlap. Jeff Davis, the author of that feature has a two part write-up on this: part 1, part 2. Depesz also has some code examples describing the feature.

Answer 2

In the meantime (since version 9.2 if I read the manual correctly) postgreSQL has added support for rangetypes.

With those rangetypes the issue suddenly becomes very simple (example copied from the manual):

CREATE TABLE reservation (
    during tsrange,
    EXCLUDE USING gist (during WITH &&)
);

And that's it. Test (also copied from the manual):

INSERT INTO reservation VALUES
    ('[2010-01-01 11:30, 2010-01-01 15:00)');

INSERT 0 1

INSERT INTO reservation VALUES
    ('[2010-01-01 14:45, 2010-01-01 15:45)');

ERROR: conflicting key value violates exclusion constraint "reservation_during_excl" DETAIL: Key (during)=(["2010-01-01 14:45:00","2010-01-01 15:45:00")) conflicts with existing key (during)=(["2010-01-01 11:30:00","2010-01-01 15:00:00")).

Answer 3

Do you really want a CHECK costraint, like mentioned in the title? That is not possible, since CHECK constrains can only work one row at a time. There might be a way to do this using triggers, though...

Answer 4

-- Implementation of a CONSTRAINT on non-overlapping datetime ranges
-- , using the Postgres rulesystem.
-- This mechanism should work for 8.4, without needing triggers.(tested on 9.0)
-- We need a shadow-table for the rangesonly to avoid recursion in the rulesystem.
-- This shadow table has a canary variable with a CONSTRAINT (value=0) on it
-- , and on changes to the basetable (that overlap with an existing interval)
-- an attempt is made to modify this variable. (which of course fails)

-- CREATE SCHEMA tmp;
DROP table tmp.dates_shadow CASCADE;
CREATE table tmp.dates_shadow
    ( time_begin timestamp with time zone
    , time_end timestamp with time zone
    , overlap_canary INTEGER NOT NULL DEFAULT '0' CHECK (overlap_canary=0)
    );
ALTER table tmp.dates_shadow
    ADD PRIMARY KEY (time_begin,time_end)
    ;

DROP table tmp.dates CASCADE;
CREATE table tmp.dates
    ( time_begin timestamp with time zone
    , time_end timestamp with time zone
    , payload varchar
    );

ALTER table tmp.dates
    ADD PRIMARY KEY (time_begin,time_end)
    ;

CREATE RULE dates_i AS
    ON INSERT TO tmp.dates
    DO ALSO (
    -- verify shadow
    UPDATE tmp.dates_shadow ds
        SET overlap_canary= 1
        WHERE (ds.time_begin, ds.time_end)
           OVERLAPS ( NEW.time_begin, NEW.time_end)
        ;
    -- insert shadow
    INSERT INTO tmp.dates_shadow (time_begin,time_end)
        VALUES (NEW.time_begin, NEW.time_end)
        ;
    );

CREATE RULE dates_d AS
    ON DELETE TO tmp.dates
    DO ALSO (
    DELETE FROM tmp.dates_shadow ds
        WHERE ds.time_begin = OLD.time_begin
        AND ds.time_end = OLD.time_end
        ;
    );

CREATE RULE dates_u AS
    ON UPDATE TO tmp.dates
    WHERE NEW.time_begin <> OLD.time_begin
    AND NEW.time_end <> OLD.time_end
    DO ALSO (
    -- delete shadow
    DELETE FROM tmp.dates_shadow ds
        WHERE ds.time_begin = OLD.time_begin
        AND ds.time_end = OLD.time_end
        ;
    -- verify shadow
    UPDATE tmp.dates_shadow ds
        SET overlap_canary= 1
        WHERE (ds.time_begin, ds.time_end)
           OVERLAPS ( NEW.time_begin, NEW.time_end)
        ;
    -- insert shadow
    INSERT INTO tmp.dates_shadow (time_begin,time_end)
        VALUES (NEW.time_begin, NEW.time_end)
        ;
    );


INSERT INTO tmp.dates(time_begin,time_end) VALUES
  ('2011-09-01', '2011-09-10')
, ('2011-09-10', '2011-09-20')
, ('2011-09-20', '2011-09-30')
    ;
SELECT * FROM tmp.dates;

EXPLAIN ANALYZE
INSERT INTO tmp.dates(time_begin,time_end) VALUES ('2011-09-30', '2011-10-04')
    ;

INSERT INTO tmp.dates(time_begin,time_end) VALUES ('2011-09-02', '2011-09-04')
    ;

SELECT * FROM tmp.dates;
SELECT * FROM tmp.dates_shadow;