Download platform/core/amazon_rivals.py from fsanyoto/loopable: direct link, hf CLI and curl.
- Browser
- Download file 84.4 kB
-
https://huggingface.co/spaces/fsanyoto/loopable/resolve/main/platform/core/amazon_rivals.py
- Command line
-
hf download hf://spaces/fsanyoto/loopable/platform/core/amazon_rivals.py
-
curl -L -o amazon_rivals.py https://huggingface.co/spaces/fsanyoto/loopable/resolve/main/platform/core/amazon_rivals.py
84.4 kB
"""Per-tenant PostgreSQL money-search rival pools and daily quotes.
Discovery reads the Amazon mirror through a cursor the caller opens: core imports no harness
module (verify_layering) and no App module (rule 13). Tables are created on the first write.
"""
from contextlib import contextmanager
from dataclasses import dataclass
from datetime import date, datetime, timedelta, timezone
import json
import re
import statistics
import time
import uuid
RIVALS = 'amazon_rivals'
DISCOVERY = 'amazon_rival_discovery'
QUOTES = 'amazon_rival_quotes'
RUNS = 'amazon_rival_runs'
WANTED = 'amazon_rival_wanted'
TABLES = (RIVALS, DISCOVERY, QUOTES, RUNS, WANTED)
#: Brand Analytics keeps the top 3 clicked ASINs of each search term.
TOP_CLICKED = 3
SOURCES = frozenset(('auto', 'added'))
CHUNK = 200
BA_TABLE = 'amzn_brand_analytics'
LISTINGS_TABLE = 'amzn_listings'
PRICING_TABLE = 'amzn_pricing'
SQP_TABLE = 'amzn_search_query_performance'
RANKS_TABLE = 'amzn_apify_rank_observations'
SNAPSHOTS_TABLE = 'amzn_competitor_snapshots'
IMAGES_TABLE = 'amzn_competitor_images'
POOL_SNAPSHOT_FIELDS = ('snapshot_date', 'brand', 'title', 'bullets', 'price', 'currency', 'prime', 'offers')
POOL_IMAGE_FIELDS = ('variant', 'link', 'width', 'height')
NO_TOP3 = "Not in any search term's top 3 clicked"
NO_WEEK = 'Brand Analytics has no dated weekly reports'
NO_MONTH = 'No completed monthly Search Query Performance report for this product yet.'
NO_MONEY = 'No searches with market money for this product yet.'
NO_KEYWORD_RANKS = "No keyword ranks for this product's searches yet."
NO_PRODUCT_TYPE = 'Product type not read yet. The next daily run reads it.'
WAITING_PRODUCT_TYPE = '{count} found, waiting for the daily read of their product type.'
NOT_OURS = 'Not an ASIN in your Amazon listings'
PARENT_ROLLS_UP = 'A Parent ASIN has no rival list. Its fields roll up its child ASINs.'
NO_POOL = 'No search report or keyword ranks for this product yet.'
ASIN_RE = re.compile(r'[A-Z0-9]{10}')
POOL_FIELDS = {
RIVALS: {'origin': ('TEXT', "'click'"), 'money_won': ('DOUBLE PRECISION', '0'),
'terms_won': ('INTEGER', '0'), 'term_rank': ('INTEGER', 'NULL'),
'eligible': ('BOOLEAN', 'TRUE')},
DISCOVERY: {'month': ('DATE', 'NULL'), 'keyword_ranks_on': ('DATE', 'NULL'),
'product_type': ('TEXT', 'NULL'), 'held': ('INTEGER', '0')},
}
RIVAL_COLUMNS = ('asin', 'rival_asin', 'source', 'removed', 'click_share', 'term', 'title', 'week',
'added_by', 'updated_at', *POOL_FIELDS[RIVALS])
DISCOVERY_COLUMNS = ('asin', 'week', 'discovered_at', 'cause', 'auto_count', *POOL_FIELDS[DISCOVERY])
QUOTE_COLUMNS = ('asin', 'quote_on', 'offer_price', 'buy_box_price', 'sales_rank', 'display_group',
'offer_cause', 'buy_box_cause', 'rank_cause', 'offers', 'currency', 'run_id', 'quoted_at')
RUN_COLUMNS = ('id', 'started_at', 'progress', 'progress_at', 'finished_at', 'asin_total', 'priced',
'ranked', 'caused', 'cause', 'counts')
WANTED_COLUMNS = ('asin', 'source', 'wanted_at')
class BackendUnavailable(RuntimeError):
"""Rival rows require PostgreSQL and never fall back to an HF document."""
class MirrorUnreadable(RuntimeError):
"""The Amazon mirror could not be read. Never read as an absent table."""
class FixtureAdapter:
"""Explicit offline fixture marker, never selected as a production fallback."""
def _now():
return datetime.now(timezone.utc).replace(microsecond=0).isoformat()
def _day(value):
if value is None:
return None
if isinstance(value, datetime):
return value.date().isoformat()
if isinstance(value, date):
return value.isoformat()
return str(value)[:10]
def asin(value, field='asin'):
"""An upper-case 10 character ASIN, or ValueError naming the field."""
text = str(value or '').strip().upper()
if not ASIN_RE.fullmatch(text):
raise ValueError(f'{field} must be a 10 character ASIN.')
return text
def _asins(values):
return sorted({asin(v) for v in values})
def _backend(rt):
if rt is None:
raise BackendUnavailable('Competitors require an explicitly bound PostgreSQL tenant runtime.')
bound = rt._store() if callable(getattr(rt, '_store', None)) else rt
from core import store_pg
if not isinstance(bound, (store_pg.PgStore, FixtureAdapter)):
raise BackendUnavailable('Competitors require PostgreSQL (STORE_BACKEND=pg). HF document storage is unsupported.')
slug = getattr(bound, 'tenant_slug', None)
if (not slug or getattr(rt, 'key', slug) != slug or getattr(rt, 'store_namespace', '')
or getattr(getattr(rt, 'tenant', None), 'key', slug) != slug):
raise ValueError('The competitors runtime and its tenant store must match.')
if isinstance(bound, FixtureAdapter):
return bound
return PgRivalStore(bound)
# --------------------------------------------------------------------------- mirror reads
def weekly_period_sql():
"""Admit connector weeks with exclusive ends and legacy weeks with inclusive ends."""
start = 'TRY_CAST("_period_start" AS TIMESTAMP)'
end = 'TRY_CAST("_period_end" AS TIMESTAMP)'
return (f"isfinite({start}) AND isfinite({end}) AND "
f"date_diff('day', CAST({start} AS DATE), CAST({end} AS DATE)) IN (6, 7)")
def _week_sql():
return 'CAST(TRY_CAST("_period_start" AS TIMESTAMP) AS DATE)'
def _columns(cur, table):
try:
return {r[0] for r in cur.execute(
'SELECT column_name FROM information_schema.columns WHERE table_name = ?', [table]).fetchall()}
except Exception as exc: # noqa: BLE001
raise MirrorUnreadable(type(exc).__name__) from exc
def _read(cur, sql, args=()):
try:
return cur.execute(sql, list(args)).fetchall()
except Exception as exc: # noqa: BLE001
raise MirrorUnreadable(type(exc).__name__) from exc
def _in(values):
return '(' + ','.join('?' for _ in values) + ')'
def newest_week(cur):
"""The newest dated Brand Analytics week start as YYYY-MM-DD, None when there is none."""
if not {'_period_start', '_period_end', 'clicked_asin', 'search_term'} <= _columns(cur, BA_TABLE):
return None
rows = _read(cur, f'SELECT max({_week_sql()}) FROM {BA_TABLE} WHERE {weekly_period_sql()}')
return _day(rows[0][0]) if rows and rows[0][0] is not None else None
def our_asins(cur, candidates=None):
"""Our ASINs from `amzn_listings.asin1`, restricted to `candidates` when given."""
if candidates is not None and not candidates:
return set()
if 'asin1' not in _columns(cur, LISTINGS_TABLE):
return set()
wanted = None if candidates is None else _asins(candidates)
where = f' AND upper(trim(CAST("asin1" AS VARCHAR))) IN {_in(wanted)}' if wanted else ''
rows = _read(cur, f'SELECT DISTINCT upper(trim(CAST("asin1" AS VARCHAR))) FROM {LISTINGS_TABLE} '
f'WHERE "asin1" IS NOT NULL{where}', wanted or ())
return {r[0] for r in rows if r[0]}
def sku_asin(cur, sku):
"""The ASIN a SKU lists under, None when the listings do not name one."""
if not {'seller_sku', 'asin1'} <= _columns(cur, LISTINGS_TABLE):
return None
rows = _read(cur, f'SELECT DISTINCT upper(trim(CAST("asin1" AS VARCHAR))) FROM {LISTINGS_TABLE} '
'WHERE trim(CAST("seller_sku" AS VARCHAR)) = ? AND "asin1" IS NOT NULL LIMIT 2',
(str(sku).strip(),))
return rows[0][0] if len(rows) == 1 and rows[0][0] else None
def children(cur, parent):
"""The child ASINs of a Parent ASIN, from the sales and traffic report."""
if not {'parent_asin', 'child_asin'} <= _columns(cur, PRICING_TABLE):
return []
rows = _read(cur, f'SELECT DISTINCT upper(trim(CAST("child_asin" AS VARCHAR))) FROM {PRICING_TABLE} '
'WHERE upper(trim(CAST("parent_asin" AS VARCHAR))) = ? '
'AND upper(trim(CAST("child_asin" AS VARCHAR))) <> ? ORDER BY 1',
(parent, parent))
return [r[0] for r in rows if r[0]]
def _money_sql(cur):
"""Newest completed monthly searches per ASIN, including zero-money searches for freshness."""
required = {'asin', 'search_query_data_search_query', 'start_date', 'end_date',
'purchase_data_total_purchase_count', 'purchase_data_total_median_purchase_price_amount'}
if not required <= _columns(cur, SQP_TABLE):
return ('SELECT NULL::VARCHAR AS asin, NULL::VARCHAR AS term, NULL::DATE AS month, '
'NULL::DOUBLE AS market_money WHERE FALSE')
return f'''
WITH reports AS (
SELECT upper(trim(CAST(asin AS VARCHAR))) AS asin,
lower(trim(CAST(search_query_data_search_query AS VARCHAR))) AS term,
TRY_CAST(start_date AS DATE) AS start_on, TRY_CAST(end_date AS DATE) AS end_on,
TRY_CAST(purchase_data_total_purchase_count AS DOUBLE) *
TRY_CAST(purchase_data_total_median_purchase_price_amount AS DOUBLE) AS market_money
FROM {SQP_TABLE}
WHERE date_diff('day', TRY_CAST(start_date AS DATE), TRY_CAST(end_date AS DATE)) > 7
AND TRY_CAST(end_date AS DATE) < CURRENT_DATE),
latest AS (
SELECT * FROM reports QUALIFY end_on = max(end_on) OVER (PARTITION BY asin))
SELECT asin, term, CAST(date_trunc('month', start_on) AS DATE) AS month,
max(market_money) AS market_money
FROM latest GROUP BY asin, term, month'''
def _click_sql(cur):
"""Normalized dated top-three clicks, with percentage shares kept in their source units."""
cols = _columns(cur, BA_TABLE)
if not {'search_term', 'clicked_asin', 'click_share_rank', 'click_share',
'_period_start', '_period_end'} <= cols:
return ('SELECT NULL::VARCHAR AS term, NULL::VARCHAR AS rival, NULL::DOUBLE AS share, '
'NULL::VARCHAR AS title, NULL::INTEGER AS term_rank, NULL::DATE AS week WHERE FALSE')
title = 'CAST(clicked_item_name AS VARCHAR)' if 'clicked_item_name' in cols else 'NULL::VARCHAR'
return f'''SELECT lower(trim(CAST(search_term AS VARCHAR))) AS term,
upper(trim(CAST(clicked_asin AS VARCHAR))) AS rival,
TRY_CAST(click_share AS DOUBLE) AS share, {title} AS title,
TRY_CAST(click_share_rank AS INTEGER) AS term_rank, {_week_sql()} AS week
FROM {BA_TABLE} WHERE {weekly_period_sql()}
AND TRY_CAST(click_share_rank AS INTEGER) BETWEEN 1 AND {TOP_CLICKED}'''
def _rank_rows_sql(cur):
"""Raw Keyword ranks observations include rivals absent from the own-placement view."""
if not {'keyword', 'asin', 'date', 'organic_rank', 'is_sponsored'} <= _columns(cur, RANKS_TABLE):
return ('SELECT NULL::VARCHAR AS term, NULL::VARCHAR AS rival, NULL::DATE AS observed_on, '
'NULL::INTEGER AS term_rank, NULL::BOOLEAN AS sponsored WHERE FALSE')
return f'''SELECT lower(trim(CAST(keyword AS VARCHAR))) AS term,
upper(trim(CAST(asin AS VARCHAR))) AS rival, TRY_CAST(date AS DATE) AS observed_on,
TRY_CAST(organic_rank AS INTEGER) AS term_rank,
TRY_CAST(is_sponsored AS BOOLEAN) AS sponsored FROM {RANKS_TABLE}'''
def _search_dates(cur, targets):
rows = _read(cur, f'''
WITH money AS ({_money_sql(cur)}), observations AS ({_rank_rows_sql(cur)})
SELECT m.asin, max(m.month), max(o.observed_on), count(DISTINCT m.term)
FILTER (WHERE m.market_money > 0)
FROM money m LEFT JOIN observations o ON o.term = m.term AND m.market_money > 0
WHERE m.asin IN {_in(targets)} GROUP BY m.asin''', targets)
return {a: dict(month=_day(m), keyword_ranks_on=_day(k), money_terms=n) for a, m, k, n in rows}
def _newest_snapshots_sql(cur, fields=('product_type', 'brand')):
if not {'asin', 'snapshot_date', 'product_type', 'brand'} <= _columns(cur, SNAPSHOTS_TABLE):
return 'SELECT NULL::VARCHAR AS asin, ' + ', '.join(f'NULL::VARCHAR AS {f}' for f in fields) + ' WHERE FALSE'
return (f'SELECT upper(trim(asin)) AS asin, {", ".join(fields)} FROM {SNAPSHOTS_TABLE} '
'QUALIFY row_number() OVER (PARTITION BY upper(trim(asin)) ORDER BY snapshot_date DESC) = 1')
def _brand(value):
return ''.join((value or '').casefold().split())
def _same_type(target_type, rival_type):
return bool(target_type and rival_type and target_type == rival_type)
def _other_brand(brand, own_brands):
return not _brand(brand) or _brand(brand) not in own_brands
@dataclass
class PoolFacts:
"""Newest ASIN facts and all historical own brands determine pool eligibility."""
latest: dict
own_brands: set
own_asins: set
def product_type(self, target):
return self.latest.get(target, {}).get('product_type')
def eligible(self, target, rival):
fact = self.latest.get(rival, {})
return (rival not in self.own_asins and _same_type(self.product_type(target), fact.get('product_type'))
and _other_brand(fact.get('brand'), self.own_brands))
def _pool_facts(cur, values, own, prefetched=None, candidate_rows=()):
latest = {r[1]: dict(product_type=r[-2], brand=r[-1]) for r in candidate_rows}
for part in _chunks(sorted(values - latest.keys())):
rows = _read(cur, f'SELECT asin, product_type, brand FROM ({_newest_snapshots_sql(cur)}) '
f'WHERE asin IN {_in(part)}', part)
latest.update({a: dict(product_type=t, brand=b) for a, t, b in rows})
brands = set()
if {'asin', 'brand'} <= _columns(cur, SNAPSHOTS_TABLE):
rows = _read(cur, f'SELECT DISTINCT s.brand FROM {SNAPSHOTS_TABLE} s '
f'JOIN {LISTINGS_TABLE} l ON upper(trim(s.asin)) = upper(trim(l.asin1))')
brands.update(_brand(row[0]) for row in rows if _brand(row[0]))
latest.update(prefetched or {})
brands.update(_brand(latest[a].get('brand')) for a in own if a in latest and _brand(latest[a].get('brand')))
return PoolFacts(latest, brands, own)
def _candidates(cur, targets, week):
"""Every non-own clicked or organic page-one ASIN on each target's monthly money searches."""
sql = f'''
WITH money AS ({_money_sql(cur)}), clicks AS ({_click_sql(cur)}),
observations AS ({_rank_rows_sql(cur)}),
newest_observations AS (
SELECT * FROM observations
QUALIFY observed_on = max(observed_on) OVER (PARTITION BY term)),
click AS (
SELECT term, rival, max(share) AS share, min(term_rank) AS term_rank, max(title) AS title
FROM clicks WHERE week = CAST(? AS DATE) GROUP BY term, rival),
page1 AS (
SELECT term, rival, min(term_rank) AS term_rank FROM newest_observations
WHERE sponsored = FALSE GROUP BY term, rival),
placements AS (
SELECT coalesce(c.term, p.term) AS term, coalesce(c.rival, p.rival) AS rival,
c.share, c.title, coalesce(c.term_rank, p.term_rank) AS term_rank,
c.term IS NOT NULL AS clicked, p.term IS NOT NULL AS page1
FROM click c FULL OUTER JOIN page1 p ON p.term = c.term AND p.rival = c.rival),
hits AS (
SELECT m.asin, p.*, m.market_money * coalesce(p.share, 0) / 100 AS money_won
FROM money m JOIN placements p ON p.term = m.term
WHERE m.asin IN {_in(targets)} AND m.market_money > 0 AND p.rival <> ''
AND p.rival NOT IN (SELECT upper(trim(CAST(asin1 AS VARCHAR)))
FROM {LISTINGS_TABLE} WHERE asin1 IS NOT NULL)),
totals AS (
SELECT asin, rival, max(share) AS share, sum(money_won) AS money_won,
count(*) FILTER (WHERE clicked AND term_rank = 1) AS terms_won,
CASE WHEN bool_or(clicked) AND bool_or(page1) THEN 'both'
WHEN bool_or(clicked) THEN 'click' ELSE 'page1' END AS origin
FROM hits GROUP BY asin, rival),
representative AS (
SELECT asin, rival, term, title, term_rank FROM hits
QUALIFY row_number() OVER (PARTITION BY asin, rival
ORDER BY money_won DESC, term_rank, term) = 1)
SELECT t.asin, t.rival, t.share, r.term, r.title, t.origin, t.money_won, t.terms_won, r.term_rank,
s.product_type, s.brand
FROM totals t JOIN representative r USING (asin, rival)
LEFT JOIN ({_newest_snapshots_sql(cur)}) s ON s.asin = t.rival
ORDER BY t.asin, t.money_won DESC, t.rival'''
return _read(cur, sql, (week, *targets))
# --------------------------------------------------------------------------- discovery
def discover(rt, cur, asins=None, *, candidate_rows=None, prefetched=None):
"""Refresh uncapped money-search pools while preserving removed and hand-added rivals."""
backend = _backend(rt)
requested = None if asins is None else _asins(asins)
targets = sorted(our_asins(cur, requested))
stamp = _now()
discovery = {}
for missing in sorted(set(requested or ()) - set(targets)):
discovery[missing] = dict(asin=missing, week=None, discovered_at=stamp, cause=NOT_OURS,
auto_count=0, month=None, keyword_ranks_on=None, product_type=None, held=0)
week = newest_week(cur) if targets else None
dates = _search_dates(cur, targets) if targets else {}
rows = (_candidates(cur, targets, week) if candidate_rows is None else candidate_rows) if targets else []
stored = backend.pool_rows(targets)
before = backend.rivals_of(targets)
values = set(targets) | {r[1] for r in rows} | {r['rival_asin'] for r in stored}
facts = _pool_facts(cur, values, our_asins(cur), prefetched, rows)
found, held = _eligible_candidates(rows, stored, facts, week, stamp)
for target in targets:
count = (len(found.get(target, ())) if facts.product_type(target) else
sum(r['source'] == 'auto' for r in stored if r['asin'] == target and not r['removed']))
meta = dates.get(target) or dict(month=None, keyword_ranks_on=None, money_terms=0)
cause = _pool_cause(facts.product_type(target), held.get(target, 0), week, meta, count)
discovery[target] = dict(asin=target, week=week, discovered_at=stamp, cause=cause, auto_count=count,
month=meta['month'], keyword_ranks_on=meta['keyword_ranks_on'],
product_type=facts.product_type(target), held=held.get(target, 0))
ready = [t for t in targets if facts.product_type(t)]
backend.replace_auto(ready, [row for t in ready for row in found.get(t, ())], list(discovery.values()))
backend.set_eligibility(stored, facts)
members = backend.rivals_of(targets)
pools = {target: len(members.get(target, ())) for target in targets}
return {'asins': len(targets), 'rivals': sum(len(v) for v in found.values()), 'week': week,
'pools': pools, 'rowsRead': len(rows), 'held': held,
'changed': tuple(t for t in targets if set(before.get(t, ())) != set(members.get(t, ()))),
'causes': {k: v['cause'] for k, v in discovery.items() if v['cause']}}
def _eligible_candidates(rows, stored, facts, week, stamp):
kept = {(r['asin'], r['rival_asin']) for r in stored if r['removed'] or r['source'] == 'added'}
candidates = {(r[0], r[1]) for r in rows} | {
(r['asin'], r['rival_asin']) for r in stored if r['source'] == 'added' and not r['removed']}
removed = {(r['asin'], r['rival_asin']) for r in stored if r['removed']}
held = {}
for target, rival in candidates - removed:
if (rival not in facts.own_asins and not facts.product_type(rival)
and _other_brand(facts.latest.get(rival, {}).get('brand'), facts.own_brands)):
held[target] = held.get(target, 0) + 1
found = {}
for target, rival, share, term, title, origin, money, terms, term_rank, _type, _name in rows:
if (target, rival) in kept or not facts.eligible(target, rival):
continue
found.setdefault(target, []).append(dict(
asin=target, rival_asin=rival, source='auto', removed=False, eligible=True,
click_share=None if share is None else float(share), term=term, title=title,
week=week, added_by=None, updated_at=stamp, origin=origin,
money_won=float(money), terms_won=int(terms), term_rank=term_rank))
return found, held
def _discovery_cause(week, meta, count):
if meta['month'] is None:
return NO_MONTH
if not meta['money_terms']:
return NO_MONEY
if meta['keyword_ranks_on'] is None:
return NO_KEYWORD_RANKS
if week is None:
return NO_WEEK
return None if count else NO_TOP3
def _pool_cause(product_type, held, week, meta, count):
if not product_type:
return NO_PRODUCT_TYPE
if held:
return WAITING_PRODUCT_TYPE.format(count=held)
return _discovery_cause(week, meta, count)
def stale(rt, cur, target):
"""True when the ASIN's discovery predates a weekly, monthly or Keyword ranks source."""
row = _backend(rt).discovery([target]).get(target)
if row is None:
return True
types = _read(cur, f'SELECT product_type FROM ({_newest_snapshots_sql(cur)}) WHERE asin = ?', [target])
product_type = types[0][0] if types else None
if row['product_type'] != product_type or (not product_type and row['cause'] != NO_PRODUCT_TYPE):
return True
newest = {'week': newest_week(cur), **(_search_dates(cur, [target]).get(target) or {})}
return any(newest.get(key) is not None and
(row.get(key) is None or row[key] < newest[key])
for key in ('week', 'month', 'keyword_ranks_on'))
# --------------------------------------------------------------------------- doors
def rivals(rt, target):
"""One ASIN's rivals: hand-added first, then money won and rank average, with discovery dates."""
target = asin(target)
backend = _backend(rt)
rows = backend.rivals(target)
found = backend.discovery([target]).get(target)
return {'asin': target, 'rivals': rows, 'removed': backend.removed_count(target),
'week': found['week'] if found else None,
'month': found['month'] if found else None,
'keywordRanksOn': found['keyword_ranks_on'] if found else None,
'discoveredAt': found['discovered_at'] if found else None,
'cause': found['cause'] if found else None,
'productType': found['product_type'] if found else None,
'held': found['held'] if found else 0}
def pool(rt, cur, target):
"""Read eligible pool rows through the caller's mirror cursor without writes or provider reads."""
target = asin(target)
backend = _backend(rt)
if cur is None:
return {**rivals(rt, target), 'rivals': [], 'competitors': [], 'dataAsOf': None, 'cause': NO_POOL}
own = our_asins(cur)
week = newest_week(cur)
sources = _candidates(cur, [target], week) if target in own else []
stored = backend.pool_rows([target])
facts = _pool_facts(cur, {target} | {r[1] for r in sources} | {r['rival_asin'] for r in stored},
own, candidate_rows=sources)
found, held = _eligible_candidates(sources, stored, facts, week, _now())
if facts.product_type(target):
rows = found.get(target, []) + [dict(r, eligible=True) for r in stored
if r['source'] == 'added' and not r['removed']
and facts.eligible(target, r['rival_asin'])]
else:
rows = [r for r in stored if r['eligible'] and not r['removed'] and r['rival_asin'] not in own
and _other_brand(facts.latest.get(r['rival_asin'], {}).get('brand'), facts.own_brands)]
if target not in own:
rows = []
averages = backend.rank_averages([r['rival_asin'] for r in rows])
rows = [{**r, **averages.get(r['rival_asin'], {'rank_average': None, 'rank_days': 0})} for r in rows]
meta = _search_dates(cur, [target]).get(target) or dict(month=None, keyword_ranks_on=None, money_terms=0)
cause = _pool_cause(facts.product_type(target), held.get(target, 0), week, meta, len(rows))
rows = sorted(rows, key=_pool_order)
snapshots = _pool_snapshots(cur, [target, *(r['rival_asin'] for r in rows)])
images = _pool_images(cur, [r['rival_asin'] for r in rows])
dates = [s['snapshot_date'] for s in snapshots.values() if s['snapshot_date'] is not None]
return {**rivals(rt, target), 'rivals': rows, 'week': week,
'month': meta['month'], 'keywordRanksOn': meta['keyword_ranks_on'],
'productType': facts.product_type(target), 'held': held.get(target, 0),
'cause': cause if target in own else NOT_OURS, 'dataAsOf': max(dates, default=None),
'competitors': [_pool_competitor(r, snapshots.get(r['rival_asin'], {}),
images.get(r['rival_asin'], [])) for r in rows]}
def _pool_snapshots(cur, values):
snapshots = {}
for part in _chunks(sorted(set(values))):
rows = _read(cur, f'SELECT * FROM ({_newest_snapshots_sql(cur, POOL_SNAPSHOT_FIELDS)}) '
f'WHERE asin IN {_in(part)}', part)
for target, *fields in rows:
snapshot = dict(zip(POOL_SNAPSHOT_FIELDS, fields))
snapshot['snapshot_date'] = _day(snapshot['snapshot_date'])
snapshot['bullets'] = json.loads(snapshot['bullets'] or '[]')
snapshots[target] = snapshot
return snapshots
def _pool_images(cur, values):
"""Resolve only the newest snapshot's variants against changed-link history."""
images = {}
if not {'asin', *POOL_IMAGE_FIELDS} <= _columns(cur, IMAGES_TABLE):
return images
for part in _chunks(values):
rows = _read(cur, f'''
SELECT s.asin, i.variant, i.link, i.width, i.height
FROM ({_newest_snapshots_sql(cur, ('image_variants',))}) s,
json_each(s.image_variants) v, {IMAGES_TABLE} i
WHERE s.asin IN {_in(part)} AND i.asin = s.asin
AND i.variant = json_extract_string(v.value, '$.variant')
AND i.link = json_extract_string(v.value, '$.link')
ORDER BY s.asin, CAST(v.key AS INTEGER)''', part)
for target, *fields in rows:
images.setdefault(target, []).append(dict(zip(POOL_IMAGE_FIELDS, fields)))
return images
def _pool_competitor(row, snapshot, images):
days = row['rank_days']
return {'asin': row['rival_asin'], 'brand': snapshot.get('brand'), 'title': snapshot.get('title'),
'origin': row['origin'], 'moneyWon': row['money_won'], 'termsWon': row['terms_won'],
'term': row['term'], 'termRank': row['term_rank'], 'bestSellerRank': row['rank_average'],
'rankDays': days, 'rankCause': f'Rank average needs 7 days of ranks. {days} so far.' if days < 7 else None,
'price': snapshot.get('price'), 'currency': snapshot.get('currency'), 'prime': snapshot.get('prime'),
'offers': snapshot.get('offers'), 'imageCount': len(images), 'images': images,
'bullets': snapshot.get('bullets', []), 'snapshotOn': snapshot.get('snapshot_date')}
def add(rt, target, rival, *, user, title=None):
"""Add or restore one hand-added rival of an ASIN."""
target, rival = asin(target), asin(rival, 'rival')
if rival == target:
raise ValueError('An ASIN cannot be its own rival.')
return _backend(rt).add(dict(asin=target, rival_asin=rival, source='added', removed=False, eligible=False,
click_share=None, term=None, title=title, week=None,
added_by=str(user or '') or None, updated_at=_now(),
origin='added', money_won=0, terms_won=0, term_rank=None))
def remove(rt, target, rival, *, user):
"""Mark one rival removed, so discovery never adds it back."""
return _backend(rt).remove(asin(target), asin(rival, 'rival'), str(user or '') or None, _now())
def rival_asins(rt):
"""Every distinct rival ASIN not removed, for the daily quotes run."""
return _backend(rt).rival_asins()
def want(rt, asins, source):
"""Register ASINs another surface needs priced by the daily run. Idempotent."""
if not isinstance(source, str) or not re.fullmatch(r'[a-z0-9_]{1,64}', source):
raise ValueError('source must be a short lower-case name.')
values = _asins(asins)
if not values:
return {'wanted': 0}
return _backend(rt).want(values, source, _now())
def wanted(rt, source=None):
"""The wanted ASINs, optionally of one source."""
return _backend(rt).wanted(source)
def put_quotes(rt, rows):
"""Upsert daily quote rows keyed (asin, quote_on). Each value without a figure carries a cause."""
clean = []
for row in rows:
item = {key: row.get(key) for key in QUOTE_COLUMNS}
item['asin'] = asin(item['asin'])
item['quote_on'] = _day(item['quote_on']) or _day(datetime.now(timezone.utc))
if item['offers'] is not None and not isinstance(item['offers'], str):
item['offers'] = json.dumps(item['offers'], separators=(',', ':'))
item['quoted_at'] = item['quoted_at'] or _now()
clean.append(item)
return _backend(rt).put_quotes(clean)
def newest_quotes(rt, asins):
"""`{asin: quote}` holding each ASIN's newest quote row. An ASIN with none is absent."""
values = _asins(asins)
return _backend(rt).newest_quotes(values) if values else {}
def start_run(rt, run_id, *, asin_total):
"""Open one quotes run row."""
return _backend(rt).start_run(str(run_id), int(asin_total), _now())
def mark_run(rt, run_id, *, progress, finished=False, counts=None, cause=None, asin_total=None):
"""Record a run's progress marker, and its finish and counts when `finished`."""
return _backend(rt).mark_run(str(run_id), int(progress), _now(), finished, counts or {}, cause,
None if asin_total is None else int(asin_total))
def runs(rt, limit=10):
"""The newest run rows first."""
return _backend(rt).runs(int(limit))
def rank_average(rt, asins):
"""Mean daily rank in each ASIN's newest display group, requiring seven of its newest 30 days."""
values = _asins(asins)
found = _backend(rt).rank_averages(values) if values else {}
return {a: (found.get(a) or {}).get('rank_average') for a in values}
def _pool_order(row):
average = row.get('rank_average')
return (-float(row.get('money_won') or 0),
average is None, average if average is not None else 0, row['rival_asin'])
# --------------------------------------------------------------------------- the daily quotes run
#: The run row gets a progress marker after each block of this many ASINs.
PROGRESS_EVERY = 100
#: One SP-API call per this many ASINs: the batch size of all three reads.
READ_BATCH = 20
#: A throttled batch is retried this many times after a backoff, then keeps its cause.
THROTTLE_RETRIES = 2
BACKOFF_SECONDS = 60.0
MAX_BACKOFF_SECONDS = 300.0
#: An unfinished run, or a finished run with a cause, is retried after this many hours.
RETRY_RUN_HOURS = 3
READS = {'offers': 'offers', 'summaries': 'Buy Box price', 'facts': 'catalog facts'}
FACT_COUNT_COLUMNS = {'title': 'title', 'product_type': 'product_type', 'image': 'image', 'rank': 'sales_rank'}
NO_MIRROR = 'The Amazon data is not on this worker yet, so no ASIN was quoted.'
NO_OWN_OFFER = 'Amazon returned no offer of yours on this ASIN.'
NO_BUY_BOX = 'Amazon returned no Buy Box offer for this ASIN.'
NO_PRICE = 'Amazon returned no price for this ASIN.'
NO_RANK = 'Amazon returned no sales rank for this ASIN.'
#: The quote fields S82-T89 compares to find a changed row.
QUOTED_FIELDS = ('offer_price', 'buy_box_price', 'sales_rank', 'display_group')
def _stamp(value):
got = datetime.fromisoformat(str(value))
return got if got.tzinfo else got.replace(tzinfo=timezone.utc)
def due(rt, *, now=None):
"""True when this tenant has no run today, or its last run stopped or failed hours ago."""
now = now or datetime.now(timezone.utc)
newest = _backend(rt).runs(1)
if not newest:
return True
row = newest[0]
if _stamp(row['started_at']).date() < now.date():
return True
if row['finished_at'] and not row['cause']:
return False
last = _stamp(row['finished_at'] or row['progress_at'] or row['started_at'])
return now - last >= timedelta(hours=RETRY_RUN_HOURS)
def unresolved(row, own):
"""The quote fields of one row with neither a value nor a cause. Empty when the row is whole."""
pairs = [('buy_box_price', 'buy_box_cause'), ('sales_rank', 'rank_cause')]
if own:
pairs.insert(0, ('offer_price', 'offer_cause'))
return [value for value, cause in pairs if row.get(value) is None and not row.get(cause)]
def _number(value):
return float(value) if isinstance(value, (int, float)) and not isinstance(value, bool) else None
def _offer_quote(response, seller_id):
"""Our offer price and the Buy Box price as two values (R40), plus every offer Amazon returned."""
payload = (((response or {}).get('body') or {}).get('payload') or {})
offers = []
for offer in payload.get('Offers') or ():
if not isinstance(offer, dict):
continue
listing = offer.get('ListingPrice') or {}
offers.append({'sellerId': offer.get('SellerId'), 'price': _number(listing.get('Amount')),
'shipping': _number((offer.get('Shipping') or {}).get('Amount')),
'currency': listing.get('CurrencyCode'), 'buyBox': bool(offer.get('IsBuyBoxWinner')),
'mine': bool(offer.get('MyOffer')) or bool(seller_id and offer.get('SellerId') == seller_id),
'fba': bool(offer.get('IsFulfilledByAmazon')), 'condition': offer.get('SubCondition')})
mine = next((o for o in offers if o['mine'] and o['price'] is not None), None)
box = next((o for o in offers if o['buyBox'] and o['price'] is not None), None)
box_price = box['price'] if box else None
currency = (mine or box or {}).get('currency')
if box_price is None:
for entry in (payload.get('Summary') or {}).get('BuyBoxPrices') or ():
amount = _number(((entry or {}).get('ListingPrice') or {}).get('Amount'))
if amount is not None and str((entry or {}).get('condition') or 'New').lower() == 'new':
box_price, currency = amount, currency or entry['ListingPrice'].get('CurrencyCode')
break
return {'offer_price': mine['price'] if mine else None, 'offer_cause': None if mine else NO_OWN_OFFER,
'buy_box_price': box_price, 'buy_box_cause': None if box_price is not None else NO_BUY_BOX,
'offers': offers, 'currency': currency}
def _summary_quote(response):
"""The Buy Box price from one getCompetitiveSummary item response."""
status = ((response or {}).get('status') or {}).get('statusCode')
body = (response or {}).get('body') or {}
if status not in (None, 200):
errors = body.get('errors') or []
reason = str((errors[0] or {}).get('message') or '')[:160] if errors else ''
return {'buy_box_price': None, 'currency': None,
'buy_box_cause': f"Amazon refused the price read ({status}{': ' + reason if reason else ''})."}
options = sorted(body.get('featuredBuyingOptions') or [],
key=lambda o: 0 if str((o or {}).get('buyingOptionType') or 'New') == 'New' else 1)
offers = []
for option in options:
for offer in (option or {}).get('segmentedFeaturedOffers') or [option]:
if not isinstance(offer, dict) or not {'sellerId', 'listingPrice'}.intersection(offer):
continue
price = _number(((offer or {}).get('listingPrice') or {}).get('amount'))
offers.append({'sellerId': offer.get('sellerId'), 'price': price,
'shipping': _number((offer.get('shippingPrice') or {}).get('amount')),
'currency': (offer.get('listingPrice') or {}).get('currencyCode'),
'buyBox': True, 'mine': False, 'fba': None,
'condition': (option or {}).get('buyingOptionType') or 'New'})
box = next((offer for offer in offers if offer['price'] is not None), None)
return {'buy_box_price': box['price'] if box else None, 'buy_box_cause': None if box else NO_BUY_BOX,
'currency': box['currency'] if box else None, 'offers': offers}
def _keyed(kind, got):
if kind == 'facts':
return {str(k).upper(): v for k, v in (got or {}).items()}
if kind == 'offers':
return {str((((r or {}).get('body') or {}).get('payload') or {}).get('ASIN') or '').upper(): r
for r in got or ()}
return {str(((r or {}).get('body') or {}).get('asin') or '').upper(): r for r in got or ()}
def _throttled(exc):
return type(exc).__name__ == 'AmazonThrottled' or getattr(exc, 'status', None) == 429
def _backoff(exc, attempt):
try:
floor = float(getattr(exc, 'retry_after', None) or 0)
except (TypeError, ValueError):
floor = 0.0
return min(MAX_BACKOFF_SECONDS, max(floor, BACKOFF_SECONDS * 2 ** attempt))
def _refused(kind, asins, exc, backoffs):
"""`{asin: cause}` for every ASIN of one refused batch. A cause never names a credential."""
label = READS[kind]
if _throttled(exc):
cause = (f'Amazon throttled the {label} read (429) after {backoffs} backoffs. '
'The next daily run reads it again.')
else:
detail = re.sub('[' + chr(0x2013) + chr(0x2014) + ']', '.', str(exc))[:160].strip()
cause = (f'Amazon refused the {label} read ({type(exc).__name__}{": " + detail if detail else ""}). '
'The next daily run reads it again.')
return {a: cause for a in asins}
def _read_batches(kind, fn, asins, *, sleep, log, counts):
"""`(answers, causes)`: answers keyed by ASIN, and a cause on each ASIN of a refused batch."""
answers, causes = {}, {}
for part in _chunks(list(asins), READ_BATCH):
attempt = 0
while True:
try:
got = _keyed(kind, fn(part))
answers.update(got)
if kind == 'facts':
for target, fact in got.items():
if fact.get('cause'):
causes[target] = fact['cause']
if '(429)' in fact['cause']:
counts['throttled'] += 1
elif 'Amazon refused' in fact['cause']:
counts['refused'] += 1
break
except Exception as exc: # noqa: BLE001
if _throttled(exc) and attempt < THROTTLE_RETRIES:
wait = _backoff(exc, attempt)
attempt += 1
counts['backoffs'] += 1
log(f'{READS[kind]} read throttled (429), backing off {wait:.0f} s')
sleep(wait)
continue
counts['throttled' if _throttled(exc) else 'refused'] += len(part)
causes.update(_refused(kind, part, exc, attempt))
break
return answers, causes
def _quote_block(block, own, readers, seller_id, quote_on, run_id, sleep, log, counts,
prefetched=None, fact_causes=None):
"""One quote row per ASIN of `block`, each with a value or a cause for every field it owns."""
mine = [a for a in block if a in own]
others = [a for a in block if a not in own]
offered, offer_causes = _read_batches('offers', readers['offers'], mine, sleep=sleep, log=log, counts=counts)
summed, sum_causes = _read_batches('summaries', readers['summaries'], others, sleep=sleep, log=log,
counts=counts)
cached = prefetched or {}
facts, rank_causes = _read_batches('facts', readers['facts'], [a for a in block if a not in cached],
sleep=sleep, log=log, counts=counts)
facts.update({a: cached[a] for a in block if a in cached})
rank_causes.update(fact_causes or {})
stamp = _now()
rows = []
for target in block:
row = dict(asin=target, quote_on=quote_on, run_id=run_id, quoted_at=stamp, offer_price=None,
offer_cause=None, buy_box_price=None, buy_box_cause=None, offers=None, currency=None)
if target in own:
if target in offer_causes:
row.update(offer_cause=offer_causes[target], buy_box_cause=offer_causes[target])
elif target in offered:
row.update(_offer_quote(offered[target], seller_id))
else:
row.update(offer_cause=NO_PRICE, buy_box_cause=NO_PRICE)
elif target in sum_causes:
row['buy_box_cause'] = sum_causes[target]
elif target in summed:
row.update(_summary_quote(summed[target]))
else:
row['buy_box_cause'] = NO_PRICE
rank = facts.get(target) or {}
value = rank.get('rank') if isinstance(rank.get('rank'), int) and not isinstance(rank.get('rank'), bool) else None
row.update(sales_rank=value, display_group=rank.get('group') if value is not None else None,
rank_cause=None if value is not None else rank_causes.get(target) or NO_RANK)
rows.append(row)
return rows, [_snapshot(row, facts.get(row['asin']) or {}, row['asin'] in own) for row in rows]
def _snapshot(quote, facts, own):
"""C3 facts with price and rank taken from the exact persisted quote row."""
return {**{key: facts.get(key) for key in ('product_type', 'brand', 'title', 'bullets',
'image_variants', 'prime', 'offers', 'rank_cat', 'image')},
'asin': quote['asin'], 'snapshot_date': quote['quote_on'],
'price': quote['offer_price' if own else 'buy_box_price'], 'currency': quote['currency'],
'sales_rank': quote['sales_rank'], 'display_group': quote['display_group']}
def _tally(counts, rows, own, snapshots):
for row in rows:
counts['priced'] += int((row['offer_price'] if row['asin'] in own else None) is not None
or row['buy_box_price'] is not None)
counts['ranked'] += int(row['sales_rank'] is not None)
counts['caused'] += int(bool(row['offer_cause'] or row['buy_box_cause'] or row['rank_cause']))
for key, field in FACT_COUNT_COLUMNS.items():
counts['facts'][key] += sum(row.get(field) is not None and row[field] != '' for row in snapshots)
def _changed(before, row):
old = before.get(row['asin'])
return old is None or any(old.get(k) != row.get(k) for k in QUOTED_FIELDS)
def _prepare_pools(rt, open_mirror, facts_reader, sleep, log, counts):
with open_mirror() as cur:
if cur is None:
return None
own = our_asins(cur)
sources = _candidates(cur, sorted(own), newest_week(cur)) if own else []
stored = _backend(rt).pool_rows(sorted(own))
values = own | {r[1] for r in sources} | {r['rival_asin'] for r in stored if not r['removed']}
known = _pool_facts(cur, values, own, candidate_rows=sources)
missing = sorted(a for a in values if not known.product_type(a))
prefetched, causes = _read_batches('facts', facts_reader, missing, sleep=sleep, log=log, counts=counts)
prefetched = {a: prefetched.get(a, {}) for a in missing}
with open_mirror() as cur:
if cur is None:
return None
found = discover(rt, cur, candidate_rows=sources, prefetched=prefetched)
return own, found, prefetched, causes
def run_quotes(rt, open_mirror, *, offers, summaries, facts, seller_id=None, run_id=None, now=None,
log=None, sleep=time.sleep):
"""Discover, read unknown types outside the mirror, filter pools, then quote with cached facts."""
_backend(rt)
log = log or (lambda _message: None)
now = now or datetime.now(timezone.utc)
quote_on = now.date().isoformat()
run_id = str(run_id or f'{quote_on}-{uuid.uuid4().hex[:8]}')
counts = dict(ours=0, rivals=0, wanted=0, asins=0, priced=0, ranked=0, caused=0, refused=0, throttled=0,
backoffs=0, discovered=0, week=None, pools={}, facts=dict.fromkeys(FACT_COUNT_COLUMNS, 0))
start_run(rt, run_id, asin_total=0)
log(f'run {run_id} started')
done = 0
try:
try:
prepared = _prepare_pools(rt, open_mirror, facts, sleep, log, counts)
if prepared is None:
return _finish(rt, run_id, done, counts, NO_MIRROR, log)
own, found, prefetched, fact_causes = prepared
except MirrorUnreadable as exc:
return _finish(rt, run_id, done, counts, f'The Amazon data could not be read ({exc}).', log)
counts.update(discovered=found['rivals'], week=found['week'], pools=found['pools'])
others = sorted(set(rival_asins(rt)) - own)
extra = sorted(set(wanted(rt)) - own - set(others))
targets = sorted(own) + others + extra
counts.update(ours=len(own), rivals=len(others), wanted=len(extra), asins=len(targets))
mark_run(rt, run_id, progress=0, counts=counts, asin_total=len(targets))
log(f'run {run_id} quotes {len(targets)} ASINs: {len(own)} ours, {len(others)} rivals, '
f'{len(extra)} wanted')
before = newest_quotes(rt, targets) if targets else {}
readers = {'offers': offers, 'summaries': summaries, 'facts': facts}
changed, snapshots = list(found['changed']), []
for block in _chunks(targets, PROGRESS_EVERY):
rows, captured = _quote_block(block, own, readers, seller_id, quote_on, run_id, sleep, log, counts,
prefetched, fact_causes)
put_quotes(rt, rows)
snapshots.extend(captured)
_tally(counts, rows, own, captured)
changed += [row['asin'] for row in rows if _changed(before, row)]
done += len(block)
mark_run(rt, run_id, progress=done, counts=counts)
log(f'run {run_id} progress {done}/{len(targets)}')
cause = (f'Amazon could not finish every read at its rate or refused a read. '
f'Retry the run within {RETRY_RUN_HOURS} hours after the rate limit delay.'
if counts['throttled'] or counts['refused'] else None)
result = _finish(rt, run_id, done, counts, cause, log)
result.update(changed=tuple(dict.fromkeys(changed)), snapshots=snapshots)
return result
except Exception as exc:
mark_run(rt, run_id, progress=done, finished=True, counts=counts,
cause=f'The run stopped ({type(exc).__name__}). It runs again within {RETRY_RUN_HOURS} hours.')
log(f'run {run_id} stopped ({type(exc).__name__}) after {done} ASINs')
raise
def _finish(rt, run_id, done, counts, cause, log):
mark_run(rt, run_id, progress=done, finished=True, counts=counts, cause=cause)
log(f'run {run_id} finished: {counts["asins"]} ASINs, {counts["priced"]} priced, {counts["ranked"]} ranked, '
f'{counts["caused"]} with a cause' + (f'. {cause}' if cause else ''))
return {'id': run_id, 'progress': done, 'cause': cause, 'counts': dict(counts), 'changed': (), 'snapshots': []}
# --------------------------------------------------------------------------- competition status and rollups
WINNING, CONTESTED, LOSING = 'Winning', 'Contested', 'Losing'
STATUSES = (WINNING, CONTESTED, LOSING)
#: The pastel green, yellow and red of the grid's brand swatches.
STATUS_COLORS = {WINNING: '#A5D8B4', CONTESTED: '#F5D989', LOSING: '#F0A8A0'}
#: R42: the status compares this price of ours to the rivals' median, never the Buy Box price.
OUR_PRICE = 'offer_price'
#: A rival's price is its featured offer, read through getCompetitiveSummary.
RIVAL_PRICE = 'buy_box_price'
GRAINS = ('sku', 'asin', 'parent_asin')
#: (key, label, type, description, extra) of the SKU and ASIN row fields, in grid order.
ROW_COLUMNS = (
('competitors', 'Competitors', 'select', 'Our offer price and sales rank against the rivals.',
{'options': list(STATUSES), 'optionColors': dict(STATUS_COLORS), 'colorCodeOptions': True}),
('competitors_cause', 'Competitors cause', 'text', 'Why the status is empty.', {}),
('rival_count', 'Rivals', 'int', 'Rivals of this ASIN.', {}),
('our_offer_price', 'Our offer price', 'currency', 'Our own offer price.', {}),
('buy_box_price', 'Buy Box price', 'currency', 'The Buy Box price on our listing.', {}),
('rival_median_price', 'Rival median price', 'currency', "The rivals' median price.", {}),
('our_sales_rank', 'Our sales rank', 'int', 'Our sales rank in our display group.', {}),
('rival_median_rank', 'Rival median rank', 'int', 'Median sales rank of rivals in our display group.', {}),
('rank_group', 'Rank group', 'text', 'The display group the ranks compare in.', {}),
)
#: R74: a Parent ASIN row carries only these rollups over its child ASINs.
PARENT_COLUMNS = (
('competitors_children', 'Children competitors', 'text', 'The worst child status, with the counts.', {}),
('children_winning', 'Children winning', 'int', 'Child ASINs that are Winning.', {}),
('children_contested', 'Children contested', 'int', 'Child ASINs that are Contested.', {}),
('children_losing', 'Children losing', 'int', 'Child ASINs that are Losing.', {}),
('children_median_offer_price', 'Children median offer price', 'currency',
'Median of our offer prices over the child ASINs.', {}),
('children_rival_median_price', 'Children rival median price', 'currency',
"Median of the child ASINs' rival median prices.", {}),
)
ROW_KEYS = tuple(c[0] for c in ROW_COLUMNS)
PARENT_KEYS = tuple(c[0] for c in PARENT_COLUMNS)
_KINDS = {c[0]: c[2] for c in ROW_COLUMNS + PARENT_COLUMNS}
NO_ASIN = 'This SKU has no ASIN in the listings.'
NO_QUOTE = 'No daily quote for this ASIN yet.'
NO_OFFER = 'We have no offer price on this ASIN.'
NO_RIVALS = 'No rival found for this ASIN.'
NO_RIVAL_PRICE = 'No rival has a price yet.'
NO_OWN_RANK = 'This ASIN has no sales rank yet.'
NO_GROUP_RIVAL = 'No rival has a sales rank in {group}.'
def _maybe_asin(value):
text = str(value or '').strip().upper()
return text if ASIN_RE.fullmatch(text) else None
def _one_line(text):
return ' '.join(str(text).split())
def _median(values):
return statistics.median(values) if values else None
def status_for(our_price, rival_price, our_rank, rival_rank):
"""R24: Winning when price and rank both beat the medians, Losing when both are worse."""
price_better, price_worse = our_price <= rival_price, our_price > rival_price
rank_better, rank_worse = our_rank < rival_rank, our_rank > rival_rank
if price_better and rank_better:
return WINNING
if price_worse and rank_worse:
return LOSING
return CONTESTED
def worst_child(statuses):
"""The worst status among the children, None when no child has one."""
present = set(statuses)
return next((s for s in (LOSING, CONTESTED, WINNING) if s in present), None)
def _standing(target, ours, quotes, count, discovery_cause):
our_price = _number((ours or {}).get(OUR_PRICE))
our_rank = (ours or {}).get('sales_rank')
group = (ours or {}).get('display_group') if our_rank is not None else None
prices = [p for p in (_number((q or {}).get(RIVAL_PRICE)) for q in quotes) if p is not None]
ranks = [q['sales_rank'] for q in quotes
if q and q.get('sales_rank') is not None and group is not None and q.get('display_group') == group]
median_price, median_rank = _median(prices), _median(ranks)
if ours is None:
cause = NO_QUOTE
elif our_price is None:
cause = ours.get('offer_cause') or NO_OFFER
elif not count:
cause = f'{NO_RIVALS[:-1]} ({discovery_cause}).' if discovery_cause else NO_RIVALS
elif median_price is None:
cause = NO_RIVAL_PRICE
elif our_rank is None:
cause = ours.get('rank_cause') or NO_OWN_RANK
elif median_rank is None:
cause = NO_GROUP_RIVAL.format(group=group or 'our display group')
else:
cause = None
return {'asin': target, 'competitors': None if cause else status_for(our_price, median_price, our_rank,
median_rank),
'competitors_cause': _one_line(cause) if cause else None, 'rival_count': count,
'our_offer_price': our_price, 'buy_box_price': _number((ours or {}).get('buy_box_price')),
'rival_median_price': median_price, 'our_sales_rank': our_rank, 'rival_median_rank': median_rank,
'rank_group': group, 'priced_rivals': len(prices), 'ranked_rivals': len(ranks),
'quote_on': (ours or {}).get('quote_on')}
def standing(rt, asins):
"""`{asin: standing}`: our quote against the newest quote of each rival, the medians and the status.
The status is empty with a one-line `competitors_cause` when our offer price, a priced rival, our
rank or a ranked rival in our display group is missing (R24, R42)."""
backend = _backend(rt)
targets = sorted({a for a in map(_maybe_asin, asins) if a})
if not targets:
return {}
listed = backend.rivals_of(targets)
found = backend.discovery(targets)
quotes = newest_quotes(rt, set(targets) | {r for rows in listed.values() for r in rows})
return {t: _standing(t, quotes.get(t), [quotes.get(r) for r in listed.get(t, ())], len(listed.get(t, ())),
(found.get(t) or {}).get('cause'))
for t in targets}
def rollup(children):
"""R74: a Parent ASIN's rollup over its children's standings, with no price, rank or status of its own."""
statuses = [c.get('competitors') for c in children if c and c.get('competitors')]
counts = {s: statuses.count(s) for s in STATUSES}
worst = worst_child(statuses)
unrated = sum(1 for c in children if not (c or {}).get('competitors'))
parts = [f'{counts[s]} {s}' for s in STATUSES if counts[s]] + ([f'{unrated} with no status'] if unrated else [])
offers = [c['our_offer_price'] for c in children if c and c.get('our_offer_price') is not None]
medians = [c['rival_median_price'] for c in children if c and c.get('rival_median_price') is not None]
return {'competitors_children': f'{worst} ({", ".join(parts)})' if worst else None,
'children_winning': counts[WINNING], 'children_contested': counts[CONTESTED],
'children_losing': counts[LOSING], 'children_median_offer_price': _median(offers),
'children_rival_median_price': _median(medians), 'children': len(children)}
def _cell(key, value):
"""A raw value as the stored cell string of its field; None leaves the cell blank."""
if value is None or value == '':
return None
kind = _KINDS[key]
if kind == 'currency':
return f'{float(value):.2f}'
if kind == 'int':
return str(int(round(float(value))))
return str(value)
def row_cells(found):
"""SKU or ASIN row cells of one standing."""
return {key: _cell(key, found.get(key)) for key in ROW_KEYS}
def parent_cells(found):
"""Parent ASIN row cells of one rollup."""
return {key: _cell(key, found.get(key)) for key in PARENT_KEYS}
def family(asin_rows):
"""`{parent: [child ASIN]}` from ASIN rows' `asin` and `parent_asin` cells. An ASIN with no parent is its own."""
out = {}
for row in asin_rows.values():
child = _maybe_asin(row.get('asin'))
if child:
out.setdefault(_maybe_asin(row.get('parent_asin')) or child, set()).add(child)
return {parent: sorted(kids) for parent, kids in out.items()}
def affected(rt, changed):
"""Our ASINs whose standing moves when these ASINs' quotes change: each one, and each ASIN they are a rival of."""
values = {a for a in map(_maybe_asin, changed or ()) if a}
return values | _backend(rt).targets_of(sorted(values)) if values else set()
def cells(rt, tables, changed=None):
"""`{grain: {row_id: {field: cell}}}` for SKU, ASIN and Parent ASIN rows; `tables` maps grain to stored rows.
With `changed` (the ASINs a quotes run changed) only the rows those ASINs move are returned."""
sku_rows, asin_rows = tables.get('sku') or {}, tables.get('asin') or {}
parent_rows = tables.get('parent_asin') or {}
kids = family(asin_rows)
hit = None if changed is None else affected(rt, changed)
def wanted(asins):
return hit is None or bool(set(asins) & hit)
pick = {'sku': {rid: _maybe_asin(r.get('asin')) for rid, r in sku_rows.items()},
'asin': {rid: _maybe_asin(r.get('asin')) for rid, r in asin_rows.items()}}
pick = {g: {rid: a for rid, a in rows.items() if (a is None and hit is None) or (a and wanted([a]))}
for g, rows in pick.items()}
parents = {}
for rid, row in parent_rows.items():
parent = _maybe_asin(row.get('parent_asin'))
if parent and wanted(kids.get(parent, ())):
parents[rid] = kids.get(parent, [])
needed = {a for rows in pick.values() for a in rows.values() if a} | {a for v in parents.values() for a in v}
found = standing(rt, needed)
out = {}
for grain, rows in pick.items():
out[grain] = {rid: row_cells(found.get(a) or ({'competitors_cause': NO_QUOTE} if a else
{'competitors_cause': NO_ASIN}))
for rid, a in rows.items()}
out['parent_asin'] = {rid: parent_cells(rollup([found.get(a) for a in v])) for rid, v in parents.items()}
return out
def _middle(values):
ordered = sorted(values)
if not ordered:
return None
half = len(ordered) // 2
return ordered[half] if len(ordered) % 2 else (ordered[half - 1] + ordered[half]) / 2
def _differs(cell, value, tolerance):
if cell is None or value is None:
return (cell is None) != (value is None)
return abs(float(cell) - float(value)) > tolerance
def validate(rt, tables, plan):
"""Recompute each rival median from the quote rows by its own query and tie it to the cells of `plan`.
`plan` is what `cells` returned for `tables`. Returns checks `{check, ok, rows, mismatches}`."""
asins = {g: {rid: _maybe_asin((tables.get(g) or {}).get(rid, {}).get('asin')) for rid in plan.get(g) or {}}
for g in ('sku', 'asin')}
kids = family(tables.get('asin') or {})
parents = {rid: kids.get(_maybe_asin((tables.get('parent_asin') or {}).get(rid, {}).get('parent_asin')), [])
for rid in plan.get('parent_asin') or {}}
targets = sorted({a for rows in asins.values() for a in rows.values() if a}
| {a for v in parents.values() for a in v})
own, rows = _backend(rt).median_inputs(targets) if targets else ({}, [])
prices, ranks = {}, {}
for target, price, rank, group in rows:
if price is not None:
prices.setdefault(target, []).append(float(price))
if rank is not None and group is not None and group == own.get(target):
ranks.setdefault(target, []).append(rank)
price = {t: _middle(prices.get(t, ())) for t in targets}
rank = {t: _middle(ranks.get(t, ())) for t in targets}
bad = {'price': [], 'rank': [], 'parent': []}
for grain, rows_ in asins.items():
for rid, target in rows_.items():
got = plan[grain][rid]
if _differs(got.get('rival_median_price'), price.get(target), 0.006):
bad['price'].append(f'{grain} {rid} {target}')
if _differs(got.get('rival_median_rank'), rank.get(target), 0.5):
bad['rank'].append(f'{grain} {rid} {target}')
for rid, children_ in parents.items():
want = _middle([price[c] for c in children_ if price.get(c) is not None])
if _differs(plan['parent_asin'][rid].get('children_rival_median_price'), want, 0.006):
bad['parent'].append(f'parent_asin {rid}')
count = sum(len(v) for v in asins.values())
return [{'check': 'rival median price ties to the quote rows', 'ok': not bad['price'], 'rows': count,
'mismatches': bad['price']},
{'check': 'rival median rank in our display group ties to the quote rows', 'ok': not bad['rank'],
'rows': count, 'mismatches': bad['rank']},
{'check': "a parent's rival median price ties to its children's", 'ok': not bad['parent'],
'rows': len(parents), 'mismatches': bad['parent']}]
# --------------------------------------------------------------------------- pool validation
def _pool_expected(cur, targets):
"""Recompute raw click money independently of discovery's candidate and money queries."""
required = {'asin', 'search_query_data_search_query', 'start_date', 'end_date',
'purchase_data_total_purchase_count', 'purchase_data_total_median_purchase_price_amount'}
clicks = {'search_term', 'clicked_asin', 'click_share', 'click_share_rank', '_period_start', '_period_end'}
if not required <= _columns(cur, SQP_TABLE) or not clicks <= _columns(cur, BA_TABLE):
return {}
rows = _read(cur, f'''
WITH monthly AS (
SELECT upper(trim(CAST(asin AS VARCHAR))) AS owner,
lower(trim(CAST(search_query_data_search_query AS VARCHAR))) AS query,
TRY_CAST(end_date AS DATE) AS ends,
TRY_CAST(purchase_data_total_purchase_count AS DOUBLE) *
TRY_CAST(purchase_data_total_median_purchase_price_amount AS DOUBLE) AS money
FROM {SQP_TABLE}
WHERE TRY_CAST(end_date AS DATE) > TRY_CAST(start_date AS DATE) + INTERVAL '7 days'
AND TRY_CAST(end_date AS DATE) < CURRENT_DATE),
newest_month AS (SELECT owner, max(ends) AS ends FROM monthly GROUP BY owner),
searches AS (
SELECT m.owner, m.query, max(m.money) AS money FROM monthly m
JOIN newest_month n ON n.owner = m.owner AND n.ends = m.ends
GROUP BY m.owner, m.query HAVING max(m.money) > 0),
report AS (
SELECT lower(trim(CAST(search_term AS VARCHAR))) AS query,
upper(trim(CAST(clicked_asin AS VARCHAR))) AS rival,
TRY_CAST(click_share AS DOUBLE) AS share,
TRY_CAST(click_share_rank AS INTEGER) AS position, {_week_sql()} AS starts
FROM {BA_TABLE} WHERE {weekly_period_sql()}),
term_clicks AS (
SELECT query, rival, max(share) AS share, min(position) AS position FROM report
WHERE starts = (SELECT max(starts) FROM report) AND position BETWEEN 1 AND 3
GROUP BY query, rival)
SELECT s.owner, c.rival, sum(s.money * coalesce(c.share, 0) * 0.01),
sum(CASE WHEN c.position = 1 THEN 1 ELSE 0 END)
FROM searches s JOIN term_clicks c ON c.query = s.query
WHERE s.owner IN {_in(targets)} GROUP BY s.owner, c.rival''', targets)
return {(a, r): (float(m), int(n)) for a, r, m, n in rows}
def validate_pool(cur, rt, targets):
"""Tie stored auto-pool money and term wins to an independent raw-source aggregation."""
targets = _asins(targets)
expected = _pool_expected(cur, targets) if targets else {}
bad_money, bad_terms, count = [], [], 0
backend = _backend(rt)
for target in targets:
for row in backend.rivals(target):
if row['source'] != 'auto':
continue
count += 1
money, terms = expected.get((target, row['rival_asin']), (0, 0))
label = f'{target} {row["rival_asin"]}'
if _differs(row['money_won'], money, 0.000001):
bad_money.append(label)
if row['terms_won'] != terms:
bad_terms.append(label)
return [{'check': 'pool money won ties to Brand Analytics and search report rows',
'ok': not bad_money, 'rows': count, 'mismatches': bad_money},
{'check': 'pool terms won ties to Brand Analytics and search report rows',
'ok': not bad_terms, 'rows': count, 'mismatches': bad_terms}]
# --------------------------------------------------------------------------- storage
def _ddl(table):
return {
RIVALS: f'''CREATE TABLE IF NOT EXISTS {table} (
tenant TEXT NOT NULL, asin TEXT NOT NULL, rival_asin TEXT NOT NULL,
source TEXT NOT NULL CHECK (source IN ('auto','added')),
removed BOOLEAN NOT NULL, click_share DOUBLE PRECISION, term TEXT, title TEXT,
week DATE, added_by TEXT, updated_at TEXT NOT NULL,
PRIMARY KEY (tenant, asin, rival_asin))''',
DISCOVERY: f'''CREATE TABLE IF NOT EXISTS {table} (
tenant TEXT NOT NULL, asin TEXT NOT NULL, week DATE, discovered_at TEXT NOT NULL,
cause TEXT, auto_count INTEGER NOT NULL, PRIMARY KEY (tenant, asin))''',
QUOTES: f'''CREATE TABLE IF NOT EXISTS {table} (
tenant TEXT NOT NULL, asin TEXT NOT NULL, quote_on DATE NOT NULL,
offer_price DOUBLE PRECISION, buy_box_price DOUBLE PRECISION, sales_rank INTEGER,
display_group TEXT, offer_cause TEXT, buy_box_cause TEXT, rank_cause TEXT, offers TEXT,
currency TEXT, run_id TEXT, quoted_at TEXT NOT NULL, PRIMARY KEY (tenant, asin, quote_on))''',
RUNS: f'''CREATE TABLE IF NOT EXISTS {table} (
tenant TEXT NOT NULL, id TEXT NOT NULL, started_at TEXT NOT NULL, progress INTEGER NOT NULL,
progress_at TEXT, finished_at TEXT, asin_total INTEGER NOT NULL, priced INTEGER,
ranked INTEGER, caused INTEGER, cause TEXT, counts TEXT, PRIMARY KEY (tenant, id))''',
WANTED: f'''CREATE TABLE IF NOT EXISTS {table} (
tenant TEXT NOT NULL, asin TEXT NOT NULL, source TEXT NOT NULL, wanted_at TEXT NOT NULL,
PRIMARY KEY (tenant, asin, source))''',
}
def _chunks(values, size=CHUNK):
for start in range(0, len(values), size):
yield values[start:start + size]
def _marks(count):
return ', '.join(['%s'] * count)
class PgRivalStore:
"""Guarded SQL storage in the schema of one explicit PgStore."""
def __init__(self, bound):
from core import store_pg
if not isinstance(bound, store_pg.PgStore):
raise TypeError('A bound PgStore is required.')
self.tenant_slug = bound.tenant_slug
self.schema = store_pg._schema_for(self.tenant_slug)
if not re.fullmatch(r'[a-zA-Z0-9_]+', self.schema):
raise ValueError('Invalid tenant schema.')
def t(self, name):
return f'"{self.schema}"."{name}"'
@contextmanager
def _connection(self):
from core import store_pg
store_pg._effective_slug(self.tenant_slug)
with store_pg._pool().connection() as con:
yield con
def _write_guard(self):
from core import store_pg
store_pg.check_write(self.tenant_slug, 'write Amazon rivals')
def _exists(self, con, name):
return con.execute('SELECT to_regclass(%s)', (self.t(name),)).fetchone()[0] is not None
def ensure_schema(self, con):
"""Create every rival table and index, only on a write."""
for name in TABLES:
con.execute(_ddl(self.t(name))[name])
for name, fields in POOL_FIELDS.items():
for column, (kind, default) in fields.items():
con.execute(f'ALTER TABLE {self.t(name)} ADD COLUMN IF NOT EXISTS {column} {kind} '
f'DEFAULT {default}')
con.execute(f'CREATE INDEX IF NOT EXISTS "amazon_rivals_rival" ON {self.t(RIVALS)} (tenant, rival_asin)')
@contextmanager
def _write(self):
self._write_guard()
with self._connection() as con:
self.ensure_schema(con)
yield con
def _select(self, con, name, columns, where, args, order='', *, include_ineligible=False):
if not self._exists(con, name):
return []
projection = columns
fields = POOL_FIELDS.get(name, {})
if fields.keys() & set(columns) or name == RIVALS:
existing = {r[0] for r in con.execute(
'SELECT column_name FROM information_schema.columns WHERE table_schema = %s AND table_name = %s',
(self.schema, name)).fetchall()}
projection = [f'{fields[c][1]} AS {c}' if c in fields and c not in existing else c for c in columns]
if name == RIVALS and 'eligible' in existing and not include_ineligible:
where = f'({where}) AND eligible = TRUE' if where else 'eligible = TRUE'
rows = con.execute(f'SELECT {", ".join(projection)} FROM {self.t(name)} WHERE tenant = %s'
f'{" AND " + where if where else ""}{order}',
(self.tenant_slug, *args)).fetchall()
return [dict(zip(columns, row)) for row in rows]
@staticmethod
def _rival(row):
return {**row, 'removed': bool(row['removed']), 'eligible': bool(row['eligible']), 'week': _day(row['week']),
'origin': 'added' if row['source'] == 'added' else row['origin'],
'click_share': None if row['click_share'] is None else float(row['click_share'])}
def pool_rows(self, targets):
"""Stored choices include hidden additions and removal tombstones."""
with self._connection() as con:
return [self._rival(row) for part in _chunks(list(targets)) for row in self._select(
con, RIVALS, RIVAL_COLUMNS, f'asin IN ({_marks(len(part))})', part,
include_ineligible=True)]
def set_eligibility(self, rows, facts):
with self._write() as con:
for row in rows:
if row['removed']:
continue
target, rival = row['asin'], row['rival_asin']
allowed = (facts.eligible(target, rival) if facts.product_type(target) else
row['eligible'] and rival not in facts.own_asins
and _other_brand(facts.latest.get(rival, {}).get('brand'), facts.own_brands))
if allowed != row['eligible']:
con.execute(f'UPDATE {self.t(RIVALS)} SET eligible = %s '
'WHERE tenant = %s AND asin = %s AND rival_asin = %s',
(allowed, self.tenant_slug, target, rival))
def replace_auto(self, targets, rows, discovery):
with self._write() as con:
for part in _chunks(targets):
con.execute(f"DELETE FROM {self.t(RIVALS)} WHERE tenant = %s AND asin IN ({_marks(len(part))}) "
"AND source = 'auto' AND removed = %s", (self.tenant_slug, *part, False))
cols = ('tenant',) + RIVAL_COLUMNS
for part in _chunks(rows, 100):
con.execute(f'INSERT INTO {self.t(RIVALS)} ({", ".join(cols)}) VALUES '
+ ', '.join(f'({_marks(len(cols))})' for _ in part)
+ ' ON CONFLICT (tenant, asin, rival_asin) DO NOTHING',
tuple(v for row in part for v in (self.tenant_slug, *(row[c] for c in RIVAL_COLUMNS))))
cols = ('tenant',) + DISCOVERY_COLUMNS
for part in _chunks(discovery, 100):
con.execute(f'INSERT INTO {self.t(DISCOVERY)} ({", ".join(cols)}) VALUES '
+ ', '.join(f'({_marks(len(cols))})' for _ in part)
+ ' ON CONFLICT (tenant, asin) DO UPDATE SET week = excluded.week, '
'discovered_at = excluded.discovered_at, cause = excluded.cause, '
'auto_count = excluded.auto_count, month = excluded.month, '
'keyword_ranks_on = excluded.keyword_ranks_on, '
'product_type = excluded.product_type, held = excluded.held',
tuple(v for row in part for v in (self.tenant_slug, *(row[c] for c in DISCOVERY_COLUMNS))))
def discovery(self, targets):
with self._connection() as con:
rows = []
for part in _chunks(list(targets)):
rows += self._select(con, DISCOVERY, DISCOVERY_COLUMNS, f'asin IN ({_marks(len(part))})', part)
return {row['asin']: {**row, **{key: _day(row[key]) for key in
('week', 'month', 'keyword_ranks_on')}} for row in rows}
def rivals(self, target):
with self._connection() as con:
rows = self._select(con, RIVALS, RIVAL_COLUMNS, 'asin = %s AND removed = %s', (target, False))
averages = self.rank_averages([row['rival_asin'] for row in rows])
enriched = [{**self._rival(row), **averages.get(row['rival_asin'],
{'rank_average': None, 'rank_days': 0})} for row in rows]
return sorted(enriched, key=lambda row: (row['source'] != 'added', *_pool_order(row)))
def rank_averages(self, values):
out = {}
with self._connection() as con:
if not self._exists(con, QUOTES):
return out
for part in _chunks(values):
rows = con.execute(f'''
WITH newest AS (
SELECT asin, max(quote_on) AS newest FROM {self.t(QUOTES)}
WHERE tenant = %s AND asin IN ({_marks(len(part))}) GROUP BY asin),
groups AS (
SELECT n.asin, n.newest, q.display_group FROM newest n
JOIN {self.t(QUOTES)} q ON q.asin = n.asin AND q.quote_on = n.newest
WHERE q.tenant = %s)
SELECT q.asin, CASE WHEN count(q.sales_rank) >= 7 THEN avg(q.sales_rank) END,
count(q.sales_rank)
FROM groups m JOIN {self.t(QUOTES)} q ON q.asin = m.asin
AND q.display_group = m.display_group AND q.quote_on >= m.newest - 29
AND q.quote_on <= m.newest
WHERE q.tenant = %s GROUP BY q.asin''',
(self.tenant_slug, *part, self.tenant_slug, self.tenant_slug)).fetchall()
out.update({a: dict(rank_average=None if avg is None else float(avg), rank_days=n)
for a, avg, n in rows})
return out
def removed_count(self, target):
with self._connection() as con:
if not self._exists(con, RIVALS):
return 0
return con.execute(f'SELECT count(*) FROM {self.t(RIVALS)} WHERE tenant = %s AND asin = %s '
'AND removed = %s', (self.tenant_slug, target, True)).fetchone()[0]
def add(self, row):
cols = ('tenant',) + RIVAL_COLUMNS
with self._write() as con:
got = con.execute(f'INSERT INTO {self.t(RIVALS)} ({", ".join(cols)}) VALUES ({_marks(len(cols))}) '
"ON CONFLICT (tenant, asin, rival_asin) DO UPDATE SET source = 'added', "
"origin = 'added', money_won = 0, terms_won = 0, term_rank = NULL, "
'removed = excluded.removed, added_by = excluded.added_by, '
'updated_at = excluded.updated_at, '
'title = coalesce(excluded.title, title) '
f'RETURNING {", ".join(RIVAL_COLUMNS)}',
(self.tenant_slug, *(row[c] for c in RIVAL_COLUMNS))).fetchone()
if not dict(zip(RIVAL_COLUMNS, got))['eligible']:
con.execute(f'UPDATE {self.t(DISCOVERY)} SET week = NULL, month = NULL, keyword_ranks_on = NULL '
'WHERE tenant = %s AND asin = %s', (self.tenant_slug, row['asin']))
return self._rival(dict(zip(RIVAL_COLUMNS, got)))
def remove(self, target, rival, user, stamp):
with self._write() as con:
got = con.execute(f'UPDATE {self.t(RIVALS)} SET removed = %s, added_by = %s, updated_at = %s '
'WHERE tenant = %s AND asin = %s AND rival_asin = %s '
f'RETURNING {", ".join(RIVAL_COLUMNS)}',
(True, user, stamp, self.tenant_slug, target, rival)).fetchone()
if got is None:
raise KeyError('Rival not found.')
return self._rival(dict(zip(RIVAL_COLUMNS, got)))
def rivals_of(self, targets):
"""`{asin: [rival ASIN]}` of the rivals not removed."""
out = {}
with self._connection() as con:
for part in _chunks(list(targets)):
for row in self._select(con, RIVALS, ('asin', 'rival_asin'),
f'asin IN ({_marks(len(part))}) AND removed = %s', (*part, False),
' ORDER BY asin, rival_asin'):
out.setdefault(row['asin'], []).append(row['rival_asin'])
return out
def targets_of(self, values):
"""Our ASINs that hold any of these ASINs as a rival not removed."""
out = set()
with self._connection() as con:
for part in _chunks(list(values)):
out |= {row['asin'] for row in self._select(
con, RIVALS, ('asin',), f'rival_asin IN ({_marks(len(part))}) AND removed = %s', (*part, False))}
return out
def median_inputs(self, targets):
"""`({asin: our display group}, [(asin, rival price, rival rank, rival group)])` from the newest quotes."""
own, rows = {}, []
newest = (f'q.quote_on = (SELECT max(n.quote_on) FROM {self.t(QUOTES)} AS n '
'WHERE n.tenant = q.tenant AND n.asin = q.asin)')
with self._connection() as con:
if not (self._exists(con, QUOTES) and self._exists(con, RIVALS)):
return own, rows
columns = {r[0] for r in con.execute(
'SELECT column_name FROM information_schema.columns WHERE table_schema = %s AND table_name = %s',
(self.schema, RIVALS)).fetchall()}
eligibility = 'AND r.eligible = TRUE ' if 'eligible' in columns else ''
for part in _chunks(list(targets)):
marks = _marks(len(part))
for target, rank, group in con.execute(
f'SELECT q.asin, q.sales_rank, q.display_group FROM {self.t(QUOTES)} AS q '
f'WHERE q.tenant = %s AND q.asin IN ({marks}) AND {newest}',
(self.tenant_slug, *part)).fetchall():
own[target] = group if rank is not None else None
rows += [tuple(r) for r in con.execute(
f'SELECT r.asin, q.{RIVAL_PRICE}, q.sales_rank, q.display_group FROM {self.t(RIVALS)} AS r '
f'JOIN {self.t(QUOTES)} AS q ON q.tenant = r.tenant AND q.asin = r.rival_asin '
f'WHERE r.tenant = %s AND r.removed = %s {eligibility}'
f'AND r.asin IN ({marks}) AND {newest}',
(self.tenant_slug, False, *part)).fetchall()]
return own, rows
def rival_asins(self):
with self._connection() as con:
rows = self._select(con, RIVALS, ('DISTINCT rival_asin',), 'removed = %s', (False,),
' ORDER BY rival_asin')
return [row['DISTINCT rival_asin'] for row in rows]
def want(self, values, source, stamp):
cols = ('tenant',) + WANTED_COLUMNS
with self._write() as con:
for part in _chunks(values):
con.execute(f'INSERT INTO {self.t(WANTED)} ({", ".join(cols)}) VALUES '
+ ', '.join(f'({_marks(len(cols))})' for _ in part)
+ ' ON CONFLICT (tenant, asin, source) DO NOTHING',
tuple(v for a in part for v in (self.tenant_slug, a, source, stamp)))
return {'wanted': len(values)}
def wanted(self, source):
with self._connection() as con:
rows = self._select(con, WANTED, ('DISTINCT asin',), 'source = %s' if source else '',
(source,) if source else (), ' ORDER BY asin')
return [row['DISTINCT asin'] for row in rows]
def put_quotes(self, rows):
cols = ('tenant',) + QUOTE_COLUMNS
update = ', '.join(f'{c} = excluded.{c}' for c in QUOTE_COLUMNS if c not in ('asin', 'quote_on'))
with self._write() as con:
for part in _chunks(rows, 100):
con.execute(f'INSERT INTO {self.t(QUOTES)} ({", ".join(cols)}) VALUES '
+ ', '.join(f'({_marks(len(cols))})' for _ in part)
+ f' ON CONFLICT (tenant, asin, quote_on) DO UPDATE SET {update}',
tuple(v for row in part for v in (self.tenant_slug, *(row[c] for c in QUOTE_COLUMNS))))
return {'quotes': len(rows)}
def newest_quotes(self, values):
out = {}
cols = ', '.join(f'q.{c}' for c in QUOTE_COLUMNS)
with self._connection() as con:
if not self._exists(con, QUOTES):
return out
for part in _chunks(values):
rows = con.execute(
f'SELECT {cols} FROM {self.t(QUOTES)} AS q JOIN ('
f'SELECT asin, max(quote_on) AS newest FROM {self.t(QUOTES)} '
f'WHERE tenant = %s AND asin IN ({_marks(len(part))}) GROUP BY asin) AS m '
'ON q.asin = m.asin AND q.quote_on = m.newest WHERE q.tenant = %s',
(self.tenant_slug, *part, self.tenant_slug)).fetchall()
for row in rows:
item = dict(zip(QUOTE_COLUMNS, row))
item['quote_on'] = _day(item['quote_on'])
item['offers'] = json.loads(item['offers']) if item['offers'] else None
out[item['asin']] = item
return out
def start_run(self, run_id, total, stamp):
with self._write() as con:
con.execute(f'INSERT INTO {self.t(RUNS)} (tenant, id, started_at, progress, asin_total) '
'VALUES (%s, %s, %s, %s, %s)', (self.tenant_slug, run_id, stamp, 0, total))
return {'id': run_id, 'started_at': stamp, 'asin_total': total}
def mark_run(self, run_id, progress, stamp, finished, counts, cause, total=None):
with self._write() as con:
got = con.execute(
f'UPDATE {self.t(RUNS)} SET progress = %s, progress_at = %s, finished_at = %s, '
'priced = %s, ranked = %s, caused = %s, cause = %s, counts = %s, '
'asin_total = coalesce(%s, asin_total) '
'WHERE tenant = %s AND id = %s RETURNING id',
(progress, stamp, stamp if finished else None, counts.get('priced'), counts.get('ranked'),
counts.get('caused'), cause, json.dumps(counts, sort_keys=True) if counts else None,
total, self.tenant_slug, run_id)).fetchone()
if got is None:
raise KeyError('Run not found.')
return {'id': run_id, 'progress': progress, 'finished': finished}
def runs(self, limit):
with self._connection() as con:
rows = self._select(con, RUNS, RUN_COLUMNS, '', (), f' ORDER BY started_at DESC, id LIMIT {int(limit)}')
for row in rows:
row['counts'] = json.loads(row['counts']) if row['counts'] else {}
return rows
"""Per-tenant PostgreSQL money-search rival pools and daily quotes.
Discovery reads the Amazon mirror through a cursor the caller opens: core imports no harness
module (verify_layering) and no App module (rule 13). Tables are created on the first write.
"""
from contextlib import contextmanager
from dataclasses import dataclass
from datetime import date, datetime, timedelta, timezone
import json
import re
import statistics
import time
import uuid
RIVALS = 'amazon_rivals'
DISCOVERY = 'amazon_rival_discovery'
QUOTES = 'amazon_rival_quotes'
RUNS = 'amazon_rival_runs'
WANTED = 'amazon_rival_wanted'
TABLES = (RIVALS, DISCOVERY, QUOTES, RUNS, WANTED)
#: Brand Analytics keeps the top 3 clicked ASINs of each search term.
TOP_CLICKED = 3
SOURCES = frozenset(('auto', 'added'))
CHUNK = 200
BA_TABLE = 'amzn_brand_analytics'
LISTINGS_TABLE = 'amzn_listings'
PRICING_TABLE = 'amzn_pricing'
SQP_TABLE = 'amzn_search_query_performance'
RANKS_TABLE = 'amzn_apify_rank_observations'
SNAPSHOTS_TABLE = 'amzn_competitor_snapshots'
IMAGES_TABLE = 'amzn_competitor_images'
POOL_SNAPSHOT_FIELDS = ('snapshot_date', 'brand', 'title', 'bullets', 'price', 'currency', 'prime', 'offers')
POOL_IMAGE_FIELDS = ('variant', 'link', 'width', 'height')
NO_TOP3 = "Not in any search term's top 3 clicked"
NO_WEEK = 'Brand Analytics has no dated weekly reports'
NO_MONTH = 'No completed monthly Search Query Performance report for this product yet.'
NO_MONEY = 'No searches with market money for this product yet.'
NO_KEYWORD_RANKS = "No keyword ranks for this product's searches yet."
NO_PRODUCT_TYPE = 'Product type not read yet. The next daily run reads it.'
WAITING_PRODUCT_TYPE = '{count} found, waiting for the daily read of their product type.'
NOT_OURS = 'Not an ASIN in your Amazon listings'
PARENT_ROLLS_UP = 'A Parent ASIN has no rival list. Its fields roll up its child ASINs.'
NO_POOL = 'No search report or keyword ranks for this product yet.'
ASIN_RE = re.compile(r'[A-Z0-9]{10}')
POOL_FIELDS = {
RIVALS: {'origin': ('TEXT', "'click'"), 'money_won': ('DOUBLE PRECISION', '0'),
'terms_won': ('INTEGER', '0'), 'term_rank': ('INTEGER', 'NULL'),
'eligible': ('BOOLEAN', 'TRUE')},
DISCOVERY: {'month': ('DATE', 'NULL'), 'keyword_ranks_on': ('DATE', 'NULL'),
'product_type': ('TEXT', 'NULL'), 'held': ('INTEGER', '0')},
}
RIVAL_COLUMNS = ('asin', 'rival_asin', 'source', 'removed', 'click_share', 'term', 'title', 'week',
'added_by', 'updated_at', *POOL_FIELDS[RIVALS])
DISCOVERY_COLUMNS = ('asin', 'week', 'discovered_at', 'cause', 'auto_count', *POOL_FIELDS[DISCOVERY])
QUOTE_COLUMNS = ('asin', 'quote_on', 'offer_price', 'buy_box_price', 'sales_rank', 'display_group',
'offer_cause', 'buy_box_cause', 'rank_cause', 'offers', 'currency', 'run_id', 'quoted_at')
RUN_COLUMNS = ('id', 'started_at', 'progress', 'progress_at', 'finished_at', 'asin_total', 'priced',
'ranked', 'caused', 'cause', 'counts')
WANTED_COLUMNS = ('asin', 'source', 'wanted_at')
class BackendUnavailable(RuntimeError):
"""Rival rows require PostgreSQL and never fall back to an HF document."""
class MirrorUnreadable(RuntimeError):
"""The Amazon mirror could not be read. Never read as an absent table."""
class FixtureAdapter:
"""Explicit offline fixture marker, never selected as a production fallback."""
def _now():
return datetime.now(timezone.utc).replace(microsecond=0).isoformat()
def _day(value):
if value is None:
return None
if isinstance(value, datetime):
return value.date().isoformat()
if isinstance(value, date):
return value.isoformat()
return str(value)[:10]
def asin(value, field='asin'):
"""An upper-case 10 character ASIN, or ValueError naming the field."""
text = str(value or '').strip().upper()
if not ASIN_RE.fullmatch(text):
raise ValueError(f'{field} must be a 10 character ASIN.')
return text