tradein-mvp/backend/data/sql/108_merge_duplicate_houses.sql

@@ -23,7 +23,7 @@
 --   house_imv_evaluations  (house_id)     — CASCADE    — UNIQUE(house_id) → dedup
 --   house_suggestions      (house_id)     — CASCADE    — UNIQUE(house_id, ext_item_id) → dedup
 --   external_valuations    (house_id)     — CASCADE    — нет ограничивающего UNIQUE
---   address_mismatch_audit (house_id)     — CASCADE    — UNIQUE(house_id,audit_batch) → dedup
+--   address_mismatch_audit (house_id)     — CASCADE    — нет ограничивающего UNIQUE
 --
 -- Идемпотентность: все UPDATE/DELETE ищут по mapping временной таблицы. При повторном
 --   запуске mapping будет пустым (дублей нет) → все UPDATE/DELETE затронут 0 строк → no-op.
@@ -80,7 +80,7 @@ clusters_with_count AS (
 ),
 -- Только дома, входящие в кластер-дубль
 dup_houses AS (
-    SELECT h.id, cl.cluster_key
+    SELECT h.id, c.cluster_key
     FROM houses h
     JOIN clustered cl ON cl.id = h.id
     JOIN clusters_with_count cw ON cw.cluster_key = cl.cluster_key
@@ -203,58 +203,47 @@ FROM _108_dup_mapping m
 WHERE haa.house_id = m.dup_id;
 
 -- 3c. houses_price_dynamics  UNIQUE(house_id, month_date, source)
---   Collision-safe: дедуп по ЦЕЛЕВОМУ ключу (canonical, month_date, source) —
---   ловит и dup-vs-canonical, и dup-vs-dup в кластерах >2. Выживает строка
---   canonical'а (m.canonical_id IS NULL), иначе минимальный id.
-DELETE FROM houses_price_dynamics t
-USING (
-    SELECT t2.id,
-           ROW_NUMBER() OVER (
-               PARTITION BY COALESCE(m.canonical_id, t2.house_id), t2.month_date, t2.source
-               ORDER BY (m.canonical_id IS NULL) DESC, t2.id ASC
-           ) AS rn
-    FROM houses_price_dynamics t2
-    LEFT JOIN _108_dup_mapping m ON m.dup_id = t2.house_id
-) d
-WHERE t.id = d.id AND d.rn > 1;
---   Репойнтить оставшиеся (целевой ключ теперь свободен)
+--   Удалить строки дубля, у которых (month_date,source) уже есть у canonical.
+DELETE FROM houses_price_dynamics hpd
+USING _108_dup_mapping m
+WHERE hpd.house_id = m.dup_id
+  AND EXISTS (
+      SELECT 1 FROM houses_price_dynamics hpd2
+      WHERE hpd2.house_id = m.canonical_id
+        AND hpd2.month_date = hpd.month_date
+        AND hpd2.source = hpd.source
+  );
+--   Репойнтить оставшиеся
 UPDATE houses_price_dynamics hpd
 SET house_id = m.canonical_id
 FROM _108_dup_mapping m
 WHERE hpd.house_id = m.dup_id;
 
--- 3d. house_imv_evaluations  UNIQUE(house_id)
---   Collision-safe: одна evaluation на canonical. Выживает canonical'а, иначе min id.
-DELETE FROM house_imv_evaluations t
-USING (
-    SELECT t2.id,
-           ROW_NUMBER() OVER (
-               PARTITION BY COALESCE(m.canonical_id, t2.house_id)
-               ORDER BY (m.canonical_id IS NULL) DESC, t2.id ASC
-           ) AS rn
-    FROM house_imv_evaluations t2
-    LEFT JOIN _108_dup_mapping m ON m.dup_id = t2.house_id
-) d
-WHERE t.id = d.id AND d.rn > 1;
---   Репойнтить оставшиеся
+-- 3d. house_imv_evaluations  UNIQUE(house_id) — у canonical уже может быть запись
+--   Удалить строки дубля если canonical уже имеет evaluation.
+DELETE FROM house_imv_evaluations hie
+USING _108_dup_mapping m
+WHERE hie.house_id = m.dup_id
+  AND EXISTS (
+      SELECT 1 FROM house_imv_evaluations hie2
+      WHERE hie2.house_id = m.canonical_id
+  );
+--   Репойнтить оставшиеся (canonical не имеет evaluation — забираем от дубля)
 UPDATE house_imv_evaluations hie
 SET house_id = m.canonical_id
 FROM _108_dup_mapping m
 WHERE hie.house_id = m.dup_id;
 
 -- 3e. house_suggestions  UNIQUE(house_id, ext_item_id)
---   Collision-safe: дедуп по (canonical, ext_item_id) — ловит dup-vs-dup.
-DELETE FROM house_suggestions t
-USING (
-    SELECT t2.id,
-           ROW_NUMBER() OVER (
-               PARTITION BY COALESCE(m.canonical_id, t2.house_id), t2.ext_item_id
-               ORDER BY (m.canonical_id IS NULL) DESC, t2.id ASC
-           ) AS rn
-    FROM house_suggestions t2
-    LEFT JOIN _108_dup_mapping m ON m.dup_id = t2.house_id
-) d
-WHERE t.id = d.id AND d.rn > 1;
+--   Удалить строки дубля, у которых (ext_item_id) уже есть у canonical.
+DELETE FROM house_suggestions hs
+USING _108_dup_mapping m
+WHERE hs.house_id = m.dup_id
+  AND EXISTS (
+      SELECT 1 FROM house_suggestions hs2
+      WHERE hs2.house_id = m.canonical_id
+        AND hs2.ext_item_id = hs.ext_item_id
+  );
 --   Репойнтить оставшиеся
 UPDATE house_suggestions hs
 SET house_id = m.canonical_id
@@ -279,20 +268,7 @@ SET house_id = m.canonical_id
 FROM _108_dup_mapping m
 WHERE ev.house_id = m.dup_id;
 
--- 3i. address_mismatch_audit  UNIQUE(house_id, audit_batch) — collision-safe
---   (заголовок ранее ошибочно помечал «нет блокирующего UNIQUE»).
-DELETE FROM address_mismatch_audit t
-USING (
-    SELECT t2.id,
-           ROW_NUMBER() OVER (
-               PARTITION BY COALESCE(m.canonical_id, t2.house_id), t2.audit_batch
-               ORDER BY (m.canonical_id IS NULL) DESC, t2.id ASC
-           ) AS rn
-    FROM address_mismatch_audit t2
-    LEFT JOIN _108_dup_mapping m ON m.dup_id = t2.house_id
-) d
-WHERE t.id = d.id AND d.rn > 1;
---   Репойнтить оставшиеся
+-- 3i. address_mismatch_audit (CASCADE, нет блокирующего UNIQUE)
 UPDATE address_mismatch_audit ama
 SET house_id = m.canonical_id
 FROM _108_dup_mapping m