safety-screen/Nominatim
1
0
Fork 0
forked from hans/Nominatim
Code Pull Requests Activity

reinstate word column in icu word table

Browse Source 
Postgresql is very bad at creating statistics for jsonb
columns. The result is that the query planer tends to
use JIT for queries with a where over 'info' even when
there is an index.
This commit is contained in:
Sarah Hoffmann
2021-07-25 15:08:11 +02:00
parent 324b1b5575
commit 1db098c05d
6 changed files with 53 additions and 65 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
lib-sql/tokenizer/icu_tokenizer_tables.sql
View File

@@ -3,6 +3,7 @@ CREATE TABLE word (
word_id INTEGER,
word_token text NOT NULL,
type text NOT NULL,
word text,
info jsonb
) {{db.tablespace.search_data}};
@@ -10,15 +11,15 @@ CREATE INDEX idx_word_word_token ON word
USING BTREE (word_token) {{db.tablespace.search_index}};
-- Used when updating country names from the boundary relation.
CREATE INDEX idx_word_country_names ON word
USING btree((info->>'cc')) {{db.tablespace.address_index}}
USING btree(word) {{db.tablespace.address_index}}
WHERE type = 'C';
-- Used when inserting new postcodes on updates.
CREATE INDEX idx_word_postcodes ON word
USING btree((info->>'postcode')) {{db.tablespace.address_index}}
USING btree(word) {{db.tablespace.address_index}}
WHERE type = 'P';
-- Used when inserting full words.
CREATE INDEX idx_word_full_word ON word
USING btree((info->>'word')) {{db.tablespace.address_index}}
USING btree(word) {{db.tablespace.address_index}}
WHERE type = 'W';
GRANT SELECT ON word TO "{{config.DATABASE_WEBUSER}}";

8
lib-sql/tokenizer/legacy_icu_tokenizer.sql
View File

@@ -98,13 +98,13 @@ DECLARE
term_count INTEGER;
BEGIN
SELECT min(word_id) INTO full_token
FROM word WHERE info->>'word' = norm_term and type = 'W';
FROM word WHERE word = norm_term and type = 'W';
IF full_token IS NULL THEN
full_token := nextval('seq_word');
INSERT INTO word (word_id, word_token, type, info)
SELECT full_token, lookup_term, 'W',
json_build_object('word', norm_term, 'count', 0)
INSERT INTO word (word_id, word_token, type, word, info)
SELECT full_token, lookup_term, 'W', norm_term,
json_build_object('count', 0)
FROM unnest(lookup_terms) as lookup_term;
END IF;