Send optimal batch size

This commit is contained in:
Juan Ignacio Sánchez Lara
2018-07-10 19:06:49 +02:00
parent 286a75fa8e
commit 531ad28158
10 changed files with 153 additions and 90 deletions

View File

@@ -112,7 +112,8 @@ CREATE TYPE cdb_dataservices_client.service_quota_info AS (
monthly_quota NUMERIC,
used_quota NUMERIC,
soft_limit BOOLEAN,
provider TEXT
provider TEXT,
max_batch_size NUMERIC
);
--
-- Public dataservices API function
@@ -1987,25 +1988,36 @@ CREATE OR REPLACE FUNCTION cdb_dataservices_client._DST_DisconnectUserTable(
TARGET cdb_dataservices_server._DST_DisconnectUserTable;
$$ LANGUAGE plproxy VOLATILE PARALLEL UNSAFE;
CREATE OR REPLACE FUNCTION cdb_dataservices_client.cdb_bulk_geocode_street_point (query text,
street_column text, city_column text default null, state_column text default null, country_column text default null, batch_size integer DEFAULT 50)
street_column text, city_column text default null, state_column text default null, country_column text default null, batch_size integer DEFAULT NULL)
RETURNS SETOF cdb_dataservices_client.geocoding AS $$
DECLARE
query_row_count integer;
enough_quota boolean;
remaining_quota integer;
max_batch_size integer;
cartodb_id_batch integer;
batches_n integer;
DEFAULT_BATCH_SIZE CONSTANT numeric := 100;
MAX_BATCH_SIZE CONSTANT numeric := 10000;
MAX_SAFE_BATCH_SIZE CONSTANT numeric := 5000;
current_row_count integer ;
temp_table_name text;
BEGIN
SELECT csqi.monthly_quota - csqi.used_quota AS remaining_quota, csqi.max_batch_size
INTO remaining_quota, max_batch_size
FROM cdb_dataservices_client.cdb_service_quota_info() csqi
WHERE service = 'hires_geocoder';
RAISE DEBUG 'remaining_quota: %; max_batch_size: %', remaining_quota, max_batch_size;
IF batch_size IS NULL THEN
RAISE EXCEPTION 'batch_size can''t be null';
ELSIF batch_size > MAX_BATCH_SIZE THEN
RAISE EXCEPTION 'batch_size must be lower than %', MAX_BATCH_SIZE + 1;
batch_size := max_batch_size;
ELSIF batch_size > max_batch_size THEN
RAISE EXCEPTION 'batch_size must be lower than %', max_batch_size + 1;
END IF;
IF batch_size > MAX_SAFE_BATCH_SIZE THEN
batch_size := MAX_SAFE_BATCH_SIZE;
END IF;
EXECUTE format('SELECT COUNT(1) from (%s) _x', query) INTO query_row_count;
@@ -2013,11 +2025,7 @@ BEGIN
RAISE DEBUG 'cdb_bulk_geocode_street_point --> query_row_count: %; query: %; country: %; state: %; city: %; street: %',
query_row_count, query, country_column, state_column, city_column, street_column;
SELECT cdb_dataservices_client.cdb_enough_quota('hires_geocoder', query_row_count) INTO enough_quota;
IF enough_quota IS NOT NULL AND NOT enough_quota THEN
SELECT csqi.monthly_quota - csqi.used_quota AS remaining_quota
INTO remaining_quota
FROM cdb_dataservices_client.cdb_service_quota_info() csqi
WHERE service = 'hires_geocoder';
IF remaining_quota < query_row_count THEN
RAISE EXCEPTION 'Remaining quota: %. Estimated cost: %', remaining_quota, query_row_count;
END IF;
@@ -2036,25 +2044,27 @@ BEGIN
coalesce(state_column, ''''''), coalesce(country_column, '''''')
into street_column, city_column, state_column, country_column;
FOR cartodb_id_batch in 0..(batches_n - 1)
LOOP
IF batches_n > 0 THEN
FOR cartodb_id_batch in 0..(batches_n - 1)
LOOP
EXECUTE format(
'WITH geocoding_data as (' ||
' SELECT ' ||
' json_build_object(''id'', cartodb_id, ''address'', %s, ''city'', %s, ''state'', %s, ''country'', %s) as data , ' ||
' floor((cartodb_id-1)::float/$1) as batch' ||
' FROM (%s) _x' ||
') ' ||
'INSERT INTO %s SELECT (cdb_dataservices_client._cdb_bulk_geocode_street_point(jsonb_agg(data))).* ' ||
'FROM geocoding_data ' ||
'WHERE batch = $2', street_column, city_column, state_column, country_column, query, temp_table_name)
USING batch_size, cartodb_id_batch;
EXECUTE format(
'WITH geocoding_data as (' ||
' SELECT ' ||
' json_build_object(''id'', cartodb_id, ''address'', %s, ''city'', %s, ''state'', %s, ''country'', %s) as data , ' ||
' floor((cartodb_id-1)::float/$1) as batch' ||
' FROM (%s) _x' ||
') ' ||
'INSERT INTO %s SELECT (cdb_dataservices_client._cdb_bulk_geocode_street_point(jsonb_agg(data))).* ' ||
'FROM geocoding_data ' ||
'WHERE batch = $2', street_column, city_column, state_column, country_column, query, temp_table_name)
USING batch_size, cartodb_id_batch;
GET DIAGNOSTICS current_row_count = ROW_COUNT;
RAISE DEBUG 'Batch % --> %', cartodb_id_batch, current_row_count;
GET DIAGNOSTICS current_row_count = ROW_COUNT;
RAISE DEBUG 'Batch % --> %', cartodb_id_batch, current_row_count;
END LOOP;
END LOOP;
END IF;
RETURN QUERY EXECUTE 'SELECT * FROM ' || quote_ident(temp_table_name);
END;

View File

@@ -39,5 +39,6 @@ CREATE TYPE cdb_dataservices_client.service_quota_info AS (
monthly_quota NUMERIC,
used_quota NUMERIC,
soft_limit BOOLEAN,
provider TEXT
provider TEXT,
max_batch_size NUMERIC
);

View File

@@ -1,23 +1,34 @@
CREATE OR REPLACE FUNCTION cdb_dataservices_client.cdb_bulk_geocode_street_point (query text,
street_column text, city_column text default null, state_column text default null, country_column text default null, batch_size integer DEFAULT 50)
street_column text, city_column text default null, state_column text default null, country_column text default null, batch_size integer DEFAULT NULL)
RETURNS SETOF cdb_dataservices_client.geocoding AS $$
DECLARE
query_row_count integer;
enough_quota boolean;
remaining_quota integer;
max_batch_size integer;
cartodb_id_batch integer;
batches_n integer;
DEFAULT_BATCH_SIZE CONSTANT numeric := 100;
MAX_BATCH_SIZE CONSTANT numeric := 10000;
MAX_SAFE_BATCH_SIZE CONSTANT numeric := 5000;
current_row_count integer ;
temp_table_name text;
BEGIN
SELECT csqi.monthly_quota - csqi.used_quota AS remaining_quota, csqi.max_batch_size
INTO remaining_quota, max_batch_size
FROM cdb_dataservices_client.cdb_service_quota_info() csqi
WHERE service = 'hires_geocoder';
RAISE DEBUG 'remaining_quota: %; max_batch_size: %', remaining_quota, max_batch_size;
IF batch_size IS NULL THEN
RAISE EXCEPTION 'batch_size can''t be null';
ELSIF batch_size > MAX_BATCH_SIZE THEN
RAISE EXCEPTION 'batch_size must be lower than %', MAX_BATCH_SIZE + 1;
batch_size := max_batch_size;
ELSIF batch_size > max_batch_size THEN
RAISE EXCEPTION 'batch_size must be lower than %', max_batch_size + 1;
END IF;
IF batch_size > MAX_SAFE_BATCH_SIZE THEN
batch_size := MAX_SAFE_BATCH_SIZE;
END IF;
EXECUTE format('SELECT COUNT(1) from (%s) _x', query) INTO query_row_count;
@@ -25,11 +36,7 @@ BEGIN
RAISE DEBUG 'cdb_bulk_geocode_street_point --> query_row_count: %; query: %; country: %; state: %; city: %; street: %',
query_row_count, query, country_column, state_column, city_column, street_column;
SELECT cdb_dataservices_client.cdb_enough_quota('hires_geocoder', query_row_count) INTO enough_quota;
IF enough_quota IS NOT NULL AND NOT enough_quota THEN
SELECT csqi.monthly_quota - csqi.used_quota AS remaining_quota
INTO remaining_quota
FROM cdb_dataservices_client.cdb_service_quota_info() csqi
WHERE service = 'hires_geocoder';
IF remaining_quota < query_row_count THEN
RAISE EXCEPTION 'Remaining quota: %. Estimated cost: %', remaining_quota, query_row_count;
END IF;
@@ -48,25 +55,27 @@ BEGIN
coalesce(state_column, ''''''), coalesce(country_column, '''''')
into street_column, city_column, state_column, country_column;
FOR cartodb_id_batch in 0..(batches_n - 1)
LOOP
IF batches_n > 0 THEN
FOR cartodb_id_batch in 0..(batches_n - 1)
LOOP
EXECUTE format(
'WITH geocoding_data as (' ||
' SELECT ' ||
' json_build_object(''id'', cartodb_id, ''address'', %s, ''city'', %s, ''state'', %s, ''country'', %s) as data , ' ||
' floor((cartodb_id-1)::float/$1) as batch' ||
' FROM (%s) _x' ||
') ' ||
'INSERT INTO %s SELECT (cdb_dataservices_client._cdb_bulk_geocode_street_point(jsonb_agg(data))).* ' ||
'FROM geocoding_data ' ||
'WHERE batch = $2', street_column, city_column, state_column, country_column, query, temp_table_name)
USING batch_size, cartodb_id_batch;
EXECUTE format(
'WITH geocoding_data as (' ||
' SELECT ' ||
' json_build_object(''id'', cartodb_id, ''address'', %s, ''city'', %s, ''state'', %s, ''country'', %s) as data , ' ||
' floor((cartodb_id-1)::float/$1) as batch' ||
' FROM (%s) _x' ||
') ' ||
'INSERT INTO %s SELECT (cdb_dataservices_client._cdb_bulk_geocode_street_point(jsonb_agg(data))).* ' ||
'FROM geocoding_data ' ||
'WHERE batch = $2', street_column, city_column, state_column, country_column, query, temp_table_name)
USING batch_size, cartodb_id_batch;
GET DIAGNOSTICS current_row_count = ROW_COUNT;
RAISE DEBUG 'Batch % --> %', cartodb_id_batch, current_row_count;
GET DIAGNOSTICS current_row_count = ROW_COUNT;
RAISE DEBUG 'Batch % --> %', cartodb_id_batch, current_row_count;
END LOOP;
END LOOP;
END IF;
RETURN QUERY EXECUTE 'SELECT * FROM ' || quote_ident(temp_table_name);
END;

View File

@@ -1,21 +1,21 @@
\set VERBOSITY terse
-- Test bulk size mandatory
SELECT cdb_dataservices_client.cdb_bulk_geocode_street_point('select 1 as cartodb_id', '''Valladolid, Spain''', null, null, null, null);
ERROR: batch_size can't be null
-- Test quota check by mocking quota 0
ALTER FUNCTION cdb_dataservices_client.cdb_service_quota_info() RENAME TO cdb_service_quota_info_mocked;
CREATE FUNCTION cdb_dataservices_client.cdb_service_quota_info ()
RETURNS SETOF cdb_dataservices_client.service_quota_info AS $$
SELECT 'hires_geocoder'::cdb_dataservices_client.service_type AS service, 0::NUMERIC AS monthly_quota, 0::NUMERIC AS used_quota, FALSE AS soft_limit, 'google' AS provider, 1::NUMERIC AS max_batch_size;
$$ LANGUAGE SQL;
ALTER FUNCTION cdb_dataservices_client.cdb_enough_quota (service TEXT ,input_size NUMERIC) RENAME TO cdb_enough_quota_mocked;
CREATE FUNCTION cdb_dataservices_client.cdb_enough_quota (service TEXT ,input_size NUMERIC)
RETURNS BOOLEAN as $$
SELECT FALSE;
$$ LANGUAGE SQL;
ALTER FUNCTION cdb_dataservices_client.cdb_service_quota_info() RENAME TO cdb_service_quota_info_mocked;
CREATE FUNCTION cdb_dataservices_client.cdb_service_quota_info ()
RETURNS SETOF cdb_dataservices_client.service_quota_info AS $$
SELECT 'hires_geocoder'::cdb_dataservices_client.service_type AS service, 0::NUMERIC AS monthly_quota, 0::NUMERIC AS used_quota, FALSE AS soft_limit, 'google' AS provider;
$$ LANGUAGE SQL;
-- Test bulk size not mandatory (it will get the optimal)
SELECT cdb_dataservices_client.cdb_bulk_geocode_street_point('select 1 as cartodb_id', '''Valladolid, Spain''', null, null, null, null);
ERROR: Remaining quota: 0. Estimated cost: 1
-- Test quota check by mocking quota 0
SELECT cdb_dataservices_client.cdb_bulk_geocode_street_point('select 1 as cartodb_id', '''Valladolid, Spain''');
ERROR: Remaining quota: 0. Estimated cost: 1
DROP FUNCTION cdb_dataservices_client.cdb_service_quota_info;
DROP FUNCTION cdb_dataservices_client.cdb_enough_quota;
ALTER FUNCTION cdb_dataservices_client.cdb_service_quota_info_mocked() RENAME TO cdb_service_quota_info;
ALTER FUNCTION cdb_dataservices_client.cdb_enough_quota_mocked (service TEXT ,input_size NUMERIC) RENAME TO cdb_enough_quota;
ALTER FUNCTION cdb_dataservices_client.cdb_service_quota_info_mocked() RENAME TO cdb_service_quota_info;

View File

@@ -1,26 +1,26 @@
\set VERBOSITY terse
-- Test bulk size mandatory
SELECT cdb_dataservices_client.cdb_bulk_geocode_street_point('select 1 as cartodb_id', '''Valladolid, Spain''', null, null, null, null);
ALTER FUNCTION cdb_dataservices_client.cdb_service_quota_info() RENAME TO cdb_service_quota_info_mocked;
CREATE FUNCTION cdb_dataservices_client.cdb_service_quota_info ()
RETURNS SETOF cdb_dataservices_client.service_quota_info AS $$
SELECT 'hires_geocoder'::cdb_dataservices_client.service_type AS service, 0::NUMERIC AS monthly_quota, 0::NUMERIC AS used_quota, FALSE AS soft_limit, 'google' AS provider, 1::NUMERIC AS max_batch_size;
$$ LANGUAGE SQL;
-- Test quota check by mocking quota 0
ALTER FUNCTION cdb_dataservices_client.cdb_enough_quota (service TEXT ,input_size NUMERIC) RENAME TO cdb_enough_quota_mocked;
CREATE FUNCTION cdb_dataservices_client.cdb_enough_quota (service TEXT ,input_size NUMERIC)
RETURNS BOOLEAN as $$
SELECT FALSE;
$$ LANGUAGE SQL;
ALTER FUNCTION cdb_dataservices_client.cdb_service_quota_info() RENAME TO cdb_service_quota_info_mocked;
CREATE FUNCTION cdb_dataservices_client.cdb_service_quota_info ()
RETURNS SETOF cdb_dataservices_client.service_quota_info AS $$
SELECT 'hires_geocoder'::cdb_dataservices_client.service_type AS service, 0::NUMERIC AS monthly_quota, 0::NUMERIC AS used_quota, FALSE AS soft_limit, 'google' AS provider;
$$ LANGUAGE SQL;
-- Test bulk size not mandatory (it will get the optimal)
SELECT cdb_dataservices_client.cdb_bulk_geocode_street_point('select 1 as cartodb_id', '''Valladolid, Spain''', null, null, null, null);
-- Test quota check by mocking quota 0
SELECT cdb_dataservices_client.cdb_bulk_geocode_street_point('select 1 as cartodb_id', '''Valladolid, Spain''');
DROP FUNCTION cdb_dataservices_client.cdb_service_quota_info;
DROP FUNCTION cdb_dataservices_client.cdb_enough_quota;
ALTER FUNCTION cdb_dataservices_client.cdb_service_quota_info_mocked() RENAME TO cdb_service_quota_info;
ALTER FUNCTION cdb_dataservices_client.cdb_enough_quota_mocked (service TEXT ,input_size NUMERIC) RENAME TO cdb_enough_quota;
ALTER FUNCTION cdb_dataservices_client.cdb_service_quota_info_mocked() RENAME TO cdb_service_quota_info;