\timing /*-------------------------------------------------------- 0. load target import to temp table 1. create pending list 2. get unqiue pending keys 3. see which keys not already in tps.trans 4. insert pending records associated with keys that are not already in trans 5. get list of recors not inserted 6. summarize records not inserted */--------------------------------------------------------- DO $$ DECLARE _t text; DECLARE _c text; BEGIN ----------------------------------------------------build the column list of the temp table---------------------------------------------------------------- SELECT string_agg(quote_ident(prs.key)||' '||prs.type,','), string_agg(quote_ident(prs.key),',') INTO _t, _c FROM TPS.srce --unwrap the schema definition array LEFT JOIN LATERAL jsonb_populate_recordset(null::tps.srce_defn_schema, defn->'schema') prs ON TRUE WHERE srce = 'HUNT' GROUP BY srce; ----------------------------------------------------add create table verbage in front of column list-------------------------------------------------------- _t := format('CREATE TEMP TABLE csv_i (%s, id SERIAL)', _t); --RAISE NOTICE '%', _t; --RAISE NOTICE '%', _c; DROP TABLE IF EXISTS csv_i; EXECUTE _t; ----------------------------------------------------do the insert------------------------------------------------------------------------------------------- --the column list needs to be dynamic forcing this whole line to be dynamic _t := format('COPY csv_i (%s) FROM ''C:\Users\fleet\downloads\hunt.csv'' WITH (HEADER TRUE,DELIMITER '','', FORMAT CSV, ENCODING ''SQL_ASCII'',QUOTE ''"'');',_c); --RAISE NOTICE '%', _t; EXECUTE _t; END $$; WITH -------------for each imported row in the COPY table, genereate the json rec, and a column for the json key specified in the srce.defn----------- pending_list AS ( SELECT ---creates a key value pair and then aggregates rows of key value pairs jsonb_object_agg( (ae.e::text[])[1], --the key name (row_to_json(i)::jsonb) #> ae.e::text[] --get the target value from the key from the csv row that has been converted to json ) json_key, row_to_json(i)::JSONB rec, srce, --ae.rn, id FROM csv_i i INNER JOIN tps.srce s ON s.srce = 'HUNT' LEFT JOIN LATERAL JSONB_ARRAY_ELEMENTS_TEXT(defn->'unique_constraint'->'fields') WITH ORDINALITY ae(e, rn) ON TRUE GROUP BY i.*, srce, id ORDER BY id ASC ) -----------create a unique list of keys from staged rows------------------------------------------------------------------------------------------ , pending_keys AS ( SELECT DISTINCT json_key FROM pending_list ) -----------return unique keys that are not already in tps.trans----------------------------------------------------------------------------------- , unmatched_keys AS ( SELECT json_key FROM pending_keys EXCEPT SELECT DISTINCT k.json_key FROM pending_keys k INNER JOIN tps.trans t ON t.rec @> k.json_key ) -----------insert pending rows that have key with no trans match----------------------------------------------------------------------------------- , inserted AS ( INSERT INTO tps.trans (srce, rec) SELECT pl.srce ,pl.rec FROM pending_list pl INNER JOIN unmatched_keys u ON u.json_key = pl.json_key ORDER BY pl.id ASC ----this conflict is only if an exact duplicate rec json happens, which will be rejected ----therefore, records may not be inserted due to ay matches with certain json fields, or if the entire json is a duplicate, reason is not specified RETURNING * ) -----------list of records not inserted-------------------------------------------------------------------------------------------------------------- , not_inserted AS ( SELECT srce ,rec FROM pending_list EXCEPT ALL SELECT srce ,rec FROM inserted ) --------summarize records not inserted-------------------+------------------------------------------------------------------------------------------------ SELECT t.srce ,(ae.e::text[])[1] unq_constr ,MIN(rec #>> ae.e::text[]) min_text ,MAX(rec #>> ae.e::text[]) max_text ,JSONB_PRETTY(JSON_AGG(rec #> ae.e::text[] ORDER BY rec #>> ae.e::text[])::JSONB) FROM not_inserted t INNER JOIN tps.srce s ON s.srce = t.srce LEFT JOIN LATERAL JSONB_ARRAY_ELEMENTS_TEXT(defn->'unique_constraint'->'fields') WITH ORDINALITY ae(e, rn) ON TRUE GROUP BY t.srce ,(ae.e::text[])[1];