PostgreSQL 9.2. Two variations of records can be stored in the following table. alarmConfigurations entries where ac_p_id is NOT NULL can exist more than once, but those where ac_p_id is NULL must only exist once (hence the ac_unique_row_no_port constraint).
As you can see, regular indexes are created on all other foreign keys, these get scanned a lot. But for ac_p_id...
- Is there any particular advantage to creating the non-unique constraint
ac_index_p_idwith IS NOT NULL (as opposed to not specifying, like the others)? - In fact, if I only create
ac_unique_row_no_port, will that also be used for scans whereac_p_idis NOT NULL?
Thanks in advance.
CREATE TABLE alarmConfigurations (
ac_id SERIAL PRIMARY KEY,
ac_ad_code TEXT NOT NULL,
ac_ad_et_code TEXT NOT NULL,
ac_e_id INTEGER NOT NULL,
ac_as_code TEXT,
ac_p_id INTEGER,
ac_details HSTORE, -- extra configuration
CONSTRAINT ac_ad_fkey FOREIGN KEY (ac_ad_code, ac_ad_et_code)
REFERENCES alarmDefinitions (ad_code, ad_et_code)
ON UPDATE CASCADE ON DELETE CASCADE,
CONSTRAINT ac_as_code_fkey FOREIGN KEY (ac_as_code)
REFERENCES alarmSeverities (as_code)
ON UPDATE CASCADE ON DELETE RESTRICT,
CONSTRAINT ac_e_id_fkey FOREIGN KEY (ac_e_id)
REFERENCES entities (e_id) ON DELETE CASCADE,
CONSTRAINT ac_p_id_fkey FOREIGN KEY (ac_p_id)
REFERENCES ports (p_id) ON DELETE CASCADE
);
CREATE INDEX ac_index_e_id ON alarmConfigurations(ac_e_id);
CREATE INDEX ac_index_ad_code ON alarmConfigurations(ac_ad_code);
CREATE INDEX ac_index_ad_et_code ON alarmConfigurations(ac_ad_et_code);
CREATE INDEX ac_index_p_id ON alarmConfigurations(ac_p_id)
WHERE ac_p_id IS NOT NULL;
CREATE UNIQUE INDEX ac_unique_row_no_port
ON alarmConfigurations(ac_ad_code, ac_ad_et_code, ac_e_id)
WHERE ac_p_id IS NULL;