fix(ods): внедрен паттерн Temporary Table для транзакционных загрузок
- Зачем: - исправление критических ошибок P0 (гонка HWM, ошибки в SQL CTE, непоследовательный lineage). - использование временных таблиц делает код более читаемым для студентов и производительным для Greenplum. - Что: - в sql/ods/ (bookings, tickets, segments, boarding_passes, flights) выборка дельты вынесена в CREATE TEMP TABLE. - HWM теперь вычисляется один раз, устраняя гонку между UPDATE и INSERT. - во всех стейтментах используется оригинальный batch_id из STG для _load_id. - поле _load_ts в ODS теперь берется из STG (load_dttm), что делает HWM-сравнение корректным. - Проверка: - визуальный аудит SQL-логики.
This commit is contained in:
@@ -1,6 +1,8 @@
|
||||
-- Загрузка ODS по boarding_passes: SCD1 (UPDATE изменившихся + INSERT новых).
|
||||
-- Используем паттерн Temporary Table для предотвращения гонки HWM между UPDATE и INSERT.
|
||||
|
||||
-- Statement 1: UPDATE существующих строк.
|
||||
-- 1. Сбор дельты во временную таблицу.
|
||||
CREATE TEMP TABLE tmp_boarding_passes_delta ON COMMIT DROP AS
|
||||
WITH src AS (
|
||||
SELECT
|
||||
s.ticket_no,
|
||||
@@ -9,23 +11,28 @@ WITH src AS (
|
||||
NULLIF(s.boarding_no, '')::INTEGER AS boarding_no,
|
||||
NULLIF(s.boarding_time, '')::TIMESTAMP WITH TIME ZONE AS boarding_time,
|
||||
s.src_created_at_ts AS event_ts,
|
||||
s.batch_id,
|
||||
s.load_dttm,
|
||||
ROW_NUMBER() OVER (
|
||||
PARTITION BY s.ticket_no, s.flight_id
|
||||
ORDER BY s.src_created_at_ts DESC NULLS LAST, s.load_dttm DESC
|
||||
) AS rn
|
||||
FROM stg.boarding_passes AS s
|
||||
-- Используем HWM (High Water Mark) по техническому времени STG
|
||||
WHERE s.load_dttm > (SELECT COALESCE(MAX(_load_ts), '1900-01-01 00:00:00'::TIMESTAMP) FROM ods.boarding_passes)
|
||||
)
|
||||
SELECT * FROM src WHERE rn = 1;
|
||||
|
||||
-- 2. UPDATE существующих строк.
|
||||
UPDATE ods.boarding_passes AS o
|
||||
SET seat_no = s.seat_no,
|
||||
boarding_no = s.boarding_no,
|
||||
boarding_time = s.boarding_time,
|
||||
event_ts = s.event_ts,
|
||||
_load_id = '{{ ti.xcom_pull(task_ids="resolve_stg_batch_id") }}'::text,
|
||||
_load_ts = now()
|
||||
FROM src AS s
|
||||
WHERE s.rn = 1
|
||||
AND o.ticket_no = s.ticket_no
|
||||
_load_id = s.batch_id, -- Сохраняем оригинальный lineage из STG
|
||||
_load_ts = s.load_dttm -- Фиксируем время STG как водяной знак для ODS
|
||||
FROM tmp_boarding_passes_delta AS s
|
||||
WHERE o.ticket_no = s.ticket_no
|
||||
AND o.flight_id = s.flight_id
|
||||
AND (
|
||||
o.seat_no IS DISTINCT FROM s.seat_no
|
||||
@@ -34,22 +41,7 @@ WHERE s.rn = 1
|
||||
OR o.event_ts IS DISTINCT FROM s.event_ts
|
||||
);
|
||||
|
||||
-- Statement 2: INSERT новых строк.
|
||||
WITH src AS (
|
||||
SELECT
|
||||
s.ticket_no,
|
||||
NULLIF(s.flight_id, '')::INTEGER AS flight_id,
|
||||
s.seat_no,
|
||||
NULLIF(s.boarding_no, '')::INTEGER AS boarding_no,
|
||||
NULLIF(s.boarding_time, '')::TIMESTAMP WITH TIME ZONE AS boarding_time,
|
||||
s.src_created_at_ts AS event_ts,
|
||||
ROW_NUMBER() OVER (
|
||||
PARTITION BY s.ticket_no, s.flight_id
|
||||
ORDER BY s.src_created_at_ts DESC NULLS LAST, s.load_dttm DESC
|
||||
) AS rn
|
||||
FROM stg.boarding_passes AS s
|
||||
WHERE s.load_dttm > (SELECT COALESCE(MAX(_load_ts), '1900-01-01 00:00:00'::TIMESTAMP) FROM ods.boarding_passes)
|
||||
)
|
||||
-- 3. INSERT новых строк.
|
||||
INSERT INTO ods.boarding_passes (
|
||||
ticket_no,
|
||||
flight_id,
|
||||
@@ -68,14 +60,13 @@ SELECT
|
||||
s.boarding_time,
|
||||
s.event_ts,
|
||||
s.batch_id,
|
||||
now()
|
||||
FROM src AS s
|
||||
WHERE s.rn = 1
|
||||
AND NOT EXISTS (
|
||||
SELECT 1
|
||||
FROM ods.boarding_passes AS o
|
||||
WHERE o.ticket_no = s.ticket_no
|
||||
AND o.flight_id = s.flight_id
|
||||
);
|
||||
s.load_dttm
|
||||
FROM tmp_boarding_passes_delta AS s
|
||||
WHERE NOT EXISTS (
|
||||
SELECT 1
|
||||
FROM ods.boarding_passes AS o
|
||||
WHERE o.ticket_no = s.ticket_no
|
||||
AND o.flight_id = s.flight_id
|
||||
);
|
||||
|
||||
ANALYZE ods.boarding_passes;
|
||||
|
||||
Reference in New Issue
Block a user