- Зачем: - текущая реализация ODS batch_id теряла данные транзакционных таблиц, если между запусками ODS STG успевал отработать дважды (брался только последний батч). - Что: - изменены скрипты загрузки транзакционных таблиц (bookings, tickets, flights, segments, boarding_passes) для использования паттерна HWM по _load_ts вместо фильтрации по конкретному батчу. - обновлен комментарий в DAG bookings_to_gp_ods, объясняющий разное поведение для справочников и транзакционных данных. - сохранено использование оригинального batch_id из STG для поля _load_id в слое ODS для сквозного трассирования. - Проверка: - запуск пайплайнов и проверка, что все батчи загружаются из STG в ODS без потерь.
82 lines
2.7 KiB
SQL
82 lines
2.7 KiB
SQL
-- Загрузка ODS по boarding_passes: SCD1 (UPDATE изменившихся + INSERT новых).
|
|
|
|
-- Statement 1: UPDATE существующих 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)
|
|
)
|
|
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
|
|
AND o.flight_id = s.flight_id
|
|
AND (
|
|
o.seat_no IS DISTINCT FROM s.seat_no
|
|
OR o.boarding_no IS DISTINCT FROM s.boarding_no
|
|
OR o.boarding_time IS DISTINCT FROM s.boarding_time
|
|
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)
|
|
)
|
|
INSERT INTO ods.boarding_passes (
|
|
ticket_no,
|
|
flight_id,
|
|
seat_no,
|
|
boarding_no,
|
|
boarding_time,
|
|
event_ts,
|
|
_load_id,
|
|
_load_ts
|
|
)
|
|
SELECT
|
|
s.ticket_no,
|
|
s.flight_id,
|
|
s.seat_no,
|
|
s.boarding_no,
|
|
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
|
|
);
|
|
|
|
ANALYZE ods.boarding_passes;
|