CREATE EXTENSION IF NOT EXISTS plpgsql WITH SCHEMA pg_catalog; CREATE EXTENSION IF NOT EXISTS tsm_system_rows; COMMENT ON EXTENSION plpgsql IS 'PL/pgSQL procedural language'; -- CREATE USER WITH ... -- createdb "gargandb" CREATE TABLE public.auth_user ( id SERIAL, password character varying(128) NOT NULL, last_login timestamp with time zone, is_superuser boolean NOT NULL, username character varying(150) NOT NULL, first_name character varying(30) NOT NULL, last_name character varying(30) NOT NULL, email character varying(254) NOT NULL, is_staff boolean NOT NULL, is_active boolean NOT NULL, date_joined timestamp with time zone DEFAULT now() NOT NULL, PRIMARY KEY (id) ); ALTER TABLE public.auth_user OWNER TO gargantua; -- TODO add publication_date -- TODO typename -> type_id CREATE TABLE public.nodes ( id SERIAL, typename integer NOT NULL, user_id integer NOT NULL, parent_id integer REFERENCES public.nodes(id) ON DELETE CASCADE , name character varying(255) DEFAULT ''::character varying NOT NULL, date timestamp with time zone DEFAULT now() NOT NULL, hyperdata jsonb DEFAULT '{}'::jsonb NOT NULL, search tsvector, PRIMARY KEY (id), FOREIGN KEY (user_id) REFERENCES public.auth_user(id) ON DELETE CASCADE ); ALTER TABLE public.nodes OWNER TO gargantua; CREATE TABLE public.ngrams ( id SERIAL, terms character varying(255), n integer, PRIMARY KEY (id) ); ALTER TABLE public.ngrams OWNER TO gargantua; -------------------------------------------------------------- -- TODO: delete delete this table --CREATE TABLE public.nodes_ngrams ( -- id SERIAL, -- node_id integer NOT NULL, -- ngrams_id integer NOT NULL, -- parent_id integer REFERENCES public.nodes_ngrams(id) ON DELETE SET NULL, -- ngrams_type integer, -- list_type integer, -- weight double precision, -- FOREIGN KEY (node_id) REFERENCES public.nodes(id) ON DELETE CASCADE, -- FOREIGN KEY (ngrams_id) REFERENCES public.ngrams(id) ON DELETE CASCADE, -- PRIMARY KEY (id) --); --ALTER TABLE public.nodes_ngrams OWNER TO gargantua; -------------------------------------------------------------- -------------------------------------------------------------- -- -- --CREATE TABLE public.nodes_ngrams_ngrams ( -- node_id integer NOT NULL REFERENCES public.nodes(id) ON DELETE CASCADE, -- ngram1_id integer NOT NULL REFERENCES public.ngrams(id) ON DELETE CASCADE, -- ngram2_id integer NOT NULL REFERENCES public.ngrams(id) ON DELETE CASCADE, -- weight double precision, -- PRIMARY KEY (node_id,ngram1_id,ngram2_id) --); -- --ALTER TABLE public.nodes_ngrams_ngrams OWNER TO gargantua; --------------------------------------------------------------- CREATE TABLE public.nodes_nodes ( node1_id integer NOT NULL REFERENCES public.nodes(id) ON DELETE CASCADE, node2_id integer NOT NULL REFERENCES public.nodes(id) ON DELETE CASCADE, score real, category integer, PRIMARY KEY (node1_id,node2_id) ); ALTER TABLE public.nodes_nodes OWNER TO gargantua; --------------------------------------------------------------- -- TODO should reference "id" of nodes_nodes (instead of node1_id, node2_id) CREATE TABLE public.node_node_ngrams ( id SERIAL, node1_id INTEGER NOT NULL REFERENCES public.nodes (id) ON DELETE CASCADE, node2_id INTEGER NOT NULL REFERENCES public.nodes (id) ON DELETE CASCADE, ngrams_id INTEGER NOT NULL REFERENCES public.ngrams (id) ON DELETE CASCADE, ngrams_type INTEGER, weight double precision, PRIMARY KEY (id) ); ALTER TABLE public.node_node_ngrams OWNER TO gargantua; -------------------------------------------------------------- --CREATE TABLE public.nodes_ngrams_repo ( -- version integer NOT NULL, -- patches jsonb DEFAULT '{}'::jsonb NOT NULL, -- PRIMARY KEY (version) --); --ALTER TABLE public.nodes_ngrams_repo OWNER TO gargantua; --------------------------------------------------------- -- If needed for rights management at row level -- CREATE EXTENSION IF NOT EXISTS acl WITH SCHEMA public; CREATE TABLE public.rights ( user_id INTEGER NOT NULL REFERENCES public.auth_user(id) ON DELETE CASCADE, node_id INTEGER NOT NULL REFERENCES public.nodes(id) ON DELETE CASCADE, rights INTEGER NOT NULL, PRIMARY KEY (user_id, node_id) ); ALTER TABLE public.rights OWNER TO gargantua; ------------------------------------------------------------ -- INDEXES CREATE INDEX ON public.auth_user USING btree (username varchar_pattern_ops); CREATE UNIQUE INDEX ON public.auth_user USING btree (username); CREATE INDEX ON public.rights USING btree (user_id,node_id); CREATE INDEX ON public.nodes USING gin (hyperdata); CREATE INDEX ON public.nodes USING btree (user_id, typename, parent_id); CREATE UNIQUE INDEX ON public.nodes USING btree (((hyperdata ->> 'uniqId'::text))); CREATE UNIQUE INDEX ON public.nodes USING btree (((hyperdata ->> 'uniqIdBdd'::text))); CREATE UNIQUE INDEX ON public.nodes USING btree (typename, parent_id, ((hyperdata ->> 'uniqId'::text))); CREATE UNIQUE INDEX ON public.ngrams (terms); -- TEST GIN CREATE INDEX ON public.nodes_nodes USING btree (node1_id, node2_id, category); CREATE UNIQUE INDEX ON public.nodes_nodes USING btree (node1_id, node2_id); CREATE UNIQUE INDEX ON public.node_node_ngrams USING btree (node1_id, node2_id, ngrams_id, ngrams_type); create INDEX on public.node_node_ngrams USING btree (node1_id, node2_id); -- TRIGGERS -- TODO user haskell-postgresql-simple to create this function -- with rights typename CREATE OR REPLACE FUNCTION public.search_update() RETURNS trigger AS $$ begin IF new.typename = 4 AND new.hyperdata @> '{"language_iso2":"EN"}' THEN new.search := to_tsvector( 'english' , (new.hyperdata ->> 'title') || ' ' || (new.hyperdata ->> 'abstract')); ELSIF new.typename = 4 AND new.hyperdata @> '{"language_iso2":"FR"}' THEN new.search := to_tsvector( 'french' , (new.hyperdata ->> 'title') || ' ' || (new.hyperdata ->> 'abstract')); ELSIF new.typename = 41 THEN new.search := to_tsvector( 'french' , (new.hyperdata ->> 'prenom') || ' ' || (new.hyperdata ->> 'nom') || ' ' || (new.hyperdata ->> 'fonction') ); ELSE new.search := to_tsvector( 'english' , new.name); END IF; return new; end $$ LANGUAGE plpgsql; ALTER FUNCTION public.search_update() OWNER TO gargantua; CREATE TRIGGER search_update_trigger BEFORE INSERT OR UPDATE ON nodes FOR EACH ROW EXECUTE PROCEDURE search_update(); -- Ngrams Full DB Extraction Optim -- TODO remove hard parameter CREATE OR REPLACE function node_pos(int, int) returns bigint AS 'SELECT count(id) from nodes WHERE id < $1 AND typename = $2 ' LANGUAGE SQL immutable; --drop index node_by_pos; create index node_by_pos on nodes using btree(node_pos(id,typename)); -- Initialize index with already existing data UPDATE nodes SET hyperdata = hyperdata;