Adds listings.card_hash (SHA-256 of volatile SERP-card fields) computed and
stored on every scrape. Skips the redundant per-day listings_snapshots write
when a listing's card is unchanged. listings.last_seen_at/is_active still
bumped on every sighting (delisting TTL). Lays groundwork for detail-refresh
gating (C2).
Подключает prod-проверенный run_cian_full_load (exhaustive региональный сбор
Cian ЕКБ вторички, room×price партиционирование, incremental on_bucket save) в
in-app scheduler как recurring-источник cian_full_load (окно 20-22 UTC — отделено
от cian_city_sweep/history_backfill на 2-5 чтобы не конкурировать на shared
kf-прокси).
run_cian_city_sweep теперь NB-only (newbuilding_only=True default): SERP-фаза
сохраняет только novostroyki-лоты, вторичку отбрасывает — ею авторитетно владеет
full_load. Убирает дубль secondary SERP-сейва. DETAIL/HOUSES-фазы не затронуты.
Backfill listings.building_cadastral_number (was 0%) from the nearest
cadastral building. gendesign_cad_buildings is a postgres_fdw foreign table
with no geom — a per-listing FDW nearest query is ~1.16s/row (~13h for 43k
listings). Instead materialize the FDW once into a LOCAL cad_buildings_local
table (Point geom + GIST), then run a fast local KNN nearest-neighbour join.
Perf: the distance gate uses geometry-space ST_DWithin(geom, point, deg) (GIST
index, no geography cast) + geom <-> KNN order + ST_DistanceSphere metric
recheck on the single nearest row. A geography-cast ST_DWithin in the WHERE
defeated the index (58s+, full update never finished); the geometry-gate design
does the whole ~41.9k-listing UPDATE in ~5.4s (refresh+match ~7s end-to-end),
matching 16027 listings at 50m across 2229 distinct buildings.
The match is GEO-NEAREST (approximate): a street-level-geocoded listing matches
the nearest building within threshold_m (default 50m), not necessarily its exact
cadastral building. Exact cadastral + parcel-containment deferred (cad_parcels
FDW not exposed). Threshold is logged.
- 124: cad_buildings_local table (empty) + GIST. Migration does not read the FDW
(deploy-independent); populated by the refresh job.
- 125: scrape_schedules seed source=cadastral_geo_match, enabled, next_run_at
tomorrow 09:00 UTC (after geocode_missing).
- tasks/cadastral_geo_match.py: refresh_cad_buildings_local (bulk FDW scan),
match_listings_to_buildings (chunked LATERAL KNN UPDATE), run_cadastral_geo_match
run-lifecycle wrapper.
- scheduler: trigger_cadastral_geo_match_run + dispatch (sync DB-only, executor).
097_index_hygiene.sql de-partialized 5 listings indexes on the false premise
that is_active=100% true. 104_index_hygiene_geom_dedup.sql acknowledged the
premise is stale (49% active on prod, 2026-06-13) but only cleaned up the
redundant geom duplicate, leaving 3 composite indexes full.
Restore WHERE is_active=true on the composite hot-path indexes:
- listings_active_filter_idx (rooms, price_rub, area_m2, scraped_at DESC)
- listings_house_price_idx (house_id_fk, price_rub)
- listings_rooms_area_idx (rooms, area_m2)
Excluded: listings_geom_active_idx (dropped by 104, listings_geom_idx covers it),
listings_scraped_desc_idx (not in issue #1398 suggested fix scope).
CONCURRENTLY not used — migration runner wraps in BEGIN/COMMIT (see 117 note).
Before #753 dedup_hash included price, so each price change created a new
active row. The canonical rows are now upserted correctly; ghost rows sit
frozen with stale hashes. Deactivates only active non-canonical rows in
groups that already have a canonical twin, excluding all-ghost groups.
Dry-run on prod (2026-06-16): UPDATE 167 rows as expected (yandex ~164,
cian ~1, n1 ~2, avito 0).
Refs #753
- New task app/tasks/avito_detail_backfill.py with run_avito_detail_backfill()
* Single snapshot SELECT at start (guarantees termination)
* Same proxy/AsyncSession path as scrape_pipeline.py step 5
* Budget guard (budget_sec), consecutive block abort (mark_done not mark_failed)
* rotate_ip() on every AvitoBlockedError; rollback on generic Exception (#1368)
* start = time.monotonic() initialized before try so except can reference it
- Scheduler wiring: trigger_avito_detail_backfill_run() + elif in scheduler_loop()
- Migration 112: scrape_schedules INSERT window 09-12 UTC, batch_size=800,
budget_sec=3600, request_delay_sec=6, max_consecutive_blocks=5
- 7 unit tests (no pytest-mock, unittest.mock only): all 7 passing,
full CI suite 1809 passed
Деплой упал на 108: (1) c.cluster_key → cl.cluster_key (несуществующий алиас);
(2) дедуп-перед-репойнтом сравнивал дубль только с каноном, но два разных дубля
одного кластера с одинаковым unique-ключом коллизили при репойнте на канон.
Переписаны 3c/houses_price_dynamics, 3d/house_imv_evaluations, 3e/house_suggestions,
3i/address_mismatch_audit на collision-safe паттерн (дедуп по целевому ключу,
канон-строка выживает). address_mismatch_audit имеет UNIQUE(house_id,audit_batch) —
заголовок ошибочно помечал «нет UNIQUE».
Валидировано dry-run против прод-данных (4478 реальных дублей):
houses 13456 → 8978, clusters_merged=3878, чисто end-to-end.
Корень проблемы: avito_houses.py обходил канонический matching-путь через
прямой INSERT ON CONFLICT(source, ext_house_id), создавая дубли для физически
одного дома (~33% таблицы houses = ~4479 лишних строк).
Forward fix: upsert_house/upsert_house_with_url → _persist_house(db, h, url),
который вызывает match_or_create_house (3-tier: source_exact→fingerprint→geo→new,
с pg_advisory_xact_lock) и затем UPDATE с enrichment-полями. avito-авторитетные
поля (rating, developer, технические хар-ки) перезаписываются напрямую;
address/lat/lon/year_built — через COALESCE чтобы не затирать canonical-данные.
Merge-миграция 108: cluster_key по cadastral/address/geo → canonical (max
не-NULL полей, geom присутствует, min(id)) → репойнт всех 11 зависимых таблиц
с обработкой UNIQUE-коллизий → DELETE дублей → backfill house_sources/aliases.
Правки уже-применённых миграций (85/062/080) не долетают до прода через
_schema_migrations. Дописаны отдельные миграции на живых таблицах:
- 158: pzz_zones_ekb UNIQUE NULLS NOT DISTINCT (rosreestr_id) (#1361)
- tradein 108: повторный backfill Avito-адресов улучшенным regexp (#1419)
- tradein 109: пересчёт asking_to_sold_ratios secondary-only (#1364/#1186)
Все три dry-run (BEGIN..ROLLBACK) против прод-схемы: PSQL_RC=0.
После того как 097 снял предикат WHERE is_active с listings_geom_active_idx,
тот стал точным дублем оригинального full listings_geom_idx (002) — оба
GIST(geom) без предиката. listings_geom_idx покрывает нагрузку (active-запросы
добавляют WHERE is_active поверх того же GIST-скана). Освобождает ~3.2 МБ.
Остальное из аудита #730 (5 dup/redundant дропов + 5 is_active де-партиалов)
уже применено миграцией 097 на prod (verify read-only подтвердил: все ABSENT
/ без предиката). Это follow-up на единственный дубль — побочный эффект 097.
Closes#730
Регистрирует отдельный in-app scheduler-источник для cian_newbuilding
enrichment-backfill (houses_price_dynamics / house_reliability_checks /
house_reviews), который раньше бежал только инлайн в Cian full-sweep.
- run_newbuilding_enrich() wrapper (heartbeat → backfill → mark_done/failed),
делегирует backfill_newbuilding_enrichment (#972, уже в main).
- trigger_newbuilding_enrich_run() + dispatch-ветка в scheduler_loop
(зеркалит trigger_sber_index_pull_run): _claim_run guard, async task,
zombie-reap наследуется; idempotency наследуется (skip уже обогащённых,
per-house SAVEPOINT, bounded per-fire limit дренирует backlog 306 домов).
- seed-миграция 103: source='newbuilding_enrich', окно 00:00-01:00 UTC
(03:00-04:00 МСК, до 01:00+ UTC sweep-блока), limit=25, next_run_at на завтра.
Засеяно enabled=FALSE (dormant): на prod весь external-HTTP scraping намеренно
на паузе. Расписание+триггер готовы, но не стартуют до намеренного возобновления
(UPDATE ... SET enabled=true). Не активирую одиночный источник при общей паузе.
Closes#973
- listings + offer_price_history SELECT для gendesign_reader
- fail-fast DO-блок: RAISE EXCEPTION если роль не существует (м.101 не применён)
- НЕ грантятся listing_sources/snapshots/events (внутренний ETL, не нужен SF)
- Контракт: consumer фильтрует vtorichka канон-предикатом сам
Closes#1191
Миграция 100_enable_deactivate_stale_avito.sql — идемпотентный UPDATE scrape_schedules SET enabled=true, next_run_at=NOW() WHERE source='deactivate_stale_avito' AND enabled=false. Чинит QA-находку #759: 090 ON CONFLICT DO NOTHING оставил pre-seed disabled-строку → task ни разу не отработала. Non-destructive, guard на повторный прогон. +7 тестов.
Closes#759.
Co-authored-by: lekss361 <lekss361@gendsgn.local>
Co-committed-by: lekss361 <lekss361@gendsgn.local>
Prod deploy 2026-05-31 13:50 failed on 095: 'cannot drop column sale_type_text
of table listings because view v_data_quality depends on it'. The view (recreated
in 094) has a CTE 'SELECT * FROM listings' which records a column-level dependency
on EVERY listings column incl sale_type_text — so the bare DROP COLUMN aborts the
whole migration (exit 1), blocking 095/096/097 from applying.
Fix: DROP VIEW v_data_quality -> DROP COLUMN sale_type_text -> recreate the view
with the identical 094 DDL (sale_type_text was never an output column, only pulled
in implicitly via SELECT *; after the drop, * no longer includes it). Verified
v_data_quality is the only object doing SELECT * FROM listings — listings_search_mv
and v_price_divergence reference explicit columns, so no other dependency blocks.
095 never applied on prod (transaction rolled back atomically, not in
_schema_migrations) → next deploy re-runs the corrected file. Editing the merged
migration in place is the correct recovery (a new 098 would not help: the runner
hits 095 first and keeps failing).
Refs #731
p3 naming-consistency (cadastral cols 0% filled). Migration 094_cadastral_unify.sql
(idempotent DO-block IF EXISTS): drop+recreate listings_search_mv (MV) +
v_data_quality (view), RENAME deals/house_metadata kadastr_num->cadastral_number,
DROP listings.kadastr_num + listings_kadastr_idx. Code SQL-string renames in
search_query/estimator/base (Lot dataclass field NOT renamed). 240 tests pass.
WARN: destructive DDL auto-applies to prod on deploy -> MERGE-GATE post-demo.
Reviewer: verify recreated MV/view DDL vs LIVE pg_get_viewdef before apply
(reconstructed from migration history 050/046, not live capture).
Refs #732
Закрывает последнюю часть #769 (A1/A2/B/C/D в #798/#804). Finding #17:
bare-city адреса геокодились в city-центроид и сохранялись как точные lat/lon →
mislocated листинги участвовали в radius ST_DWithin-аналогах.
- 089: ALTER TABLE listings ADD COLUMN IF NOT EXISTS geo_precision text (idempot).
- ScrapedLot.geo_precision + проброс в save_listings.
- geocode_missing.py: geo_precision='city' через существующий _geocode_is_coarse.
- estimator _fetch_analogs Tier H+W: + AND (geo_precision IS DISTINCT FROM 'city').
NULL проходит (консервативно, сдвига до backfill нет). Pricing не тронут.
7 тестов; 270 passed, ruff clean.
