Download platform/modules/competitor_watch.py from fsanyoto/loopable: direct link, hf CLI and curl.
- Browser
- Download file 87.4 kB
-
https://huggingface.co/spaces/fsanyoto/loopable/resolve/main/platform/modules/competitor_watch.py
- Command line
-
hf download hf://spaces/fsanyoto/loopable/platform/modules/competitor_watch.py
-
curl -L -o competitor_watch.py https://huggingface.co/spaces/fsanyoto/loopable/resolve/main/platform/modules/competitor_watch.py
87.4 kB
"""Competitor watch App: watched terms, weekly comparisons and owned seller alerts.
Store documents (tenant runtime store, App owned, never the mirror and never Amazon):
competitor_watch_settings {undercut_pct, rank_surge_pct, default_owner}
competitor_watch_terms {seeded: [...], added: [...], removed: [...], seeded_at}
competitor_watch_alerts legacy source, retained for explicit resumable migration
competitor_watch_alert/<id> one alert per store row
competitor_watch_check/<id> one check per store row
competitor_watch_alerts_index migration checkpoint and serialized alert sequence
competitor_watch_weeks {last_compared, comparison: {...}, metrics: {period_start: {asin: metric}}}
competitor_watch_terms_index {version: 2, activeTotal, trackingReady, trackingCause}, prepared by sync
competitor_watch_terms SQL table (tenant, term, label, active, starred, ours)
competitor_watch_term/<term> legacy rows, retained after one-time migration
Top 3 history is read from the connector table `amzn_brand_analytics`, which the store's weekly
retention keeps for watched terms. App documents never change the connector (rule 13).
Alerts and checks page in the tenant's existing Postgres store. No collection document or row cap
limits their growth. GET never prepares rows or reads the legacy collection.
"""
import contextlib
import datetime as dt
import json
import math
import re
from pathlib import Path
from urllib.parse import unquote_plus
from core.amazon_rivals import weekly_period_sql as _weekly_period_sql
TRIGGERS = {
"new_asin": "New ASIN in top 3",
"lost_spot": "Lost spot",
"undercut": "Undercut",
"rank_surge": "Rank surge",
"own_asin_seller": "Seller on my ASIN",
}
STATUSES = ("Open", "Repriced", "Ads changed", "Listing improved", "Watching", "Ignore", "Reported to Amazon")
#: A key (term, asin, trigger) is held while its latest alert has one of these statuses.
HOLDING = ("Open", "Watching", "Ignore", "Reported to Amazon")
DEFAULTS = {"undercut_pct": 5.0, "rank_surge_pct": 50.0}
DUE_DAYS = 7
NOT_READY = "Comparison starts after the next weekly pull."
#: The mirror holds the tables but this read failed or was busy. `{cause}` is the error class.
READ_FAILED = ("The Amazon mirror could not be read ({cause}), so no search term was compared "
"and the next Amazon sync will retry.")
BA_DATES_MISSING = "Brand Analytics report dates are missing"
BA_WEEKLY_MISSING = "Brand Analytics has no dated weekly reports"
#: Stored sync failures that a live probe finding a dated weekly report proves superseded.
SUPERSEDED_BY_WEEKLY = frozenset(READ_FAILED.format(cause=cause) for cause in (BA_DATES_MISSING, BA_WEEKLY_MISSING))
LIMIT = "Alerts and checks use separate store rows and server pages, with no row cap."
KEY_SETTINGS = "competitor_watch_settings"
KEY_TERMS = "competitor_watch_terms"
KEY_ALERTS = "competitor_watch_alerts"
KEY_WEEKS = "competitor_watch_weeks"
KEY_TERM_INDEX = "competitor_watch_terms_index"
TERM_PREFIX = "competitor_watch_term/"
TERM_TABLE = "competitor_watch_terms"
TERMS_NOT_READY = "Search terms are not ready. The next Amazon sync prepares term paging."
TRACKING_NOT_READY = "Tracked terms have no checked ownership source. The next Amazon sync will retry."
KEY_ALERT_INDEX = "competitor_watch_alerts_index"
ALERT_PREFIX = "competitor_watch_alert/"
CHECK_PREFIX = "competitor_watch_check/"
ALERTS_NOT_READY = "Alert paging is not ready. The next Amazon sync or an administrator must finish alert migration."
BA_TABLE = "amzn_brand_analytics"
SQP_TABLE = "amzn_search_query_performance"
LISTINGS_TABLE = "amzn_listings"
WAIT_QUOTES = "Waits for the daily price job"
def _now():
return dt.datetime.now(dt.timezone.utc)
def _stamp(at):
return at.strftime("%Y-%m-%dT%H:%M:%SZ")
def _norm(term):
return " ".join(str(term or "").split()).lower()
def display_term(term):
"""Decode a label without changing the stored parent-record key."""
return " ".join(unquote_plus(str(term or "")).split())
def _rt(tenant_key, st=None):
"""The tenant runtime store handle (get and update), or raise when it cannot be resolved."""
if st is not None:
return st
from harness import amazon_store
rt = amazon_store._runtime_for(tenant_key)
if rt is None:
raise RuntimeError(f"The store for tenant {tenant_key!r} could not be opened.")
return rt
def _doc(st, key):
value = st.get(key)
return value if isinstance(value, dict) else {}
class TermsNotReady(RuntimeError):
"""Term paging has not been prepared by a completed sync."""
def __init__(self, message, *, problems=()):
super().__init__(message)
self.problems = list(problems)
class AlertsNotReady(RuntimeError):
"""Alert rows are not ready for reads or edits."""
def _bound_store(tenant_key, st):
backend = st._document_store() if hasattr(st, "_document_store") else st
slug = getattr(backend, "tenant_slug", None)
if slug is not None and slug != tenant_key:
raise RuntimeError("The store handle belongs to another tenant.")
return backend
def _paths(tenant_key, st, name, paths):
backend = _bound_store(tenant_key, st)
reader = getattr(backend, "get_json_paths", None)
if callable(reader):
key = st.store_key(name) if hasattr(st, "store_key") else name
return reader(key, paths)
doc = _doc(st, name)
out = []
for path in paths:
value = doc
for part in path:
value = value.get(part) if isinstance(value, dict) else None
out.append(value)
return out
class _DocumentIndex:
"""Bind legacy document SQL to the caller's tenant."""
def __init__(self, tenant_key, st):
from core import store_pg
self.pg, self.slug, self.st = store_pg, tenant_key, st
self.key = st.store_key if hasattr(st, "store_key") else str
self.prefix = self.key(TERM_PREFIX)
def connection(self):
self.pg._effective_slug(self.slug)
return self.pg._pool().connection()
def sql(self, statement):
return self.pg._q(statement, "store_kv", self.slug)
class TermStoreFixture:
"""Explicit offline term-store adapter, never a production fallback."""
class _TermIndex(_DocumentIndex):
"""Indexed tenant term rows, prepared only by an explicit sync write."""
def __init__(self, tenant_key, st):
super().__init__(tenant_key, st)
schema = self.pg._schema_for(self.slug)
if not re.fullmatch(r"[a-zA-Z0-9_]+", schema):
raise ValueError("Invalid tenant schema.")
self.table = f'"{schema}"."{TERM_TABLE}"'
def term_sql(self, statement):
return statement.replace("{terms}", self.table)
def ready(self):
with self.connection() as con:
row = con.execute(self.sql("SELECT value ->> 'version' FROM {tbl} WHERE key = %s"),
[self.key(KEY_TERM_INDEX)]).fetchone()
return bool(row and row[0] == "2")
def legacy_all(self):
"""Keep the wider pre-migration watched list readable by retention without migrating it."""
with self.connection() as con:
row = con.execute(self.sql("SELECT value ->> 'version' FROM {tbl} WHERE key = %s"),
[self.key(KEY_TERM_INDEX)]).fetchone()
if not row or row[0] != "1":
return None
return [row[0] for row in con.execute(self.sql("SELECT value ->> 'term' FROM {tbl} "
"WHERE starts_with(key, %s) AND (value ->> 'active') = 'true'"), [self.prefix]).fetchall()]
def ensure_schema(self, con):
con.execute(self.term_sql('CREATE TABLE IF NOT EXISTS {terms} ('
'tenant TEXT NOT NULL, term TEXT COLLATE "C" NOT NULL, label TEXT NOT NULL, '
'active BOOLEAN NOT NULL DEFAULT TRUE, starred BOOLEAN NOT NULL DEFAULT FALSE, '
'ours BOOLEAN NOT NULL DEFAULT FALSE, PRIMARY KEY (tenant, term))'))
for suffix, columns in (("active", "tenant, active, term"),
("starred", "tenant, starred, term"), ("ours", "tenant, ours, term"),
("label", "tenant, active, label, term")):
con.execute(self.term_sql(f'CREATE INDEX IF NOT EXISTS "cw_terms_{suffix}" '
f'ON {{terms}} ({columns})'))
con.execute(self.term_sql('CREATE INDEX IF NOT EXISTS "cw_terms_tracked_label" '
'ON {terms} (tenant, label, term)'))
def _put_meta(self, con, meta):
con.execute(self.sql("INSERT INTO {tbl} (key, value) VALUES (%s, %s::jsonb) "
"ON CONFLICT (key) DO UPDATE SET value = excluded.value"),
[self.key(KEY_TERM_INDEX), json.dumps(meta)])
@contextlib.contextmanager
def _locked(self, *, prepare=False):
self.pg.check_write(self.slug, "write competitor watch term rows")
with self.connection() as con:
con.execute("SET LOCAL lock_timeout = '10s'")
con.execute("SET LOCAL statement_timeout = '30s'")
if prepare:
con.execute(self.sql("INSERT INTO {tbl} (key, value) VALUES (%s, %s::jsonb) "
"ON CONFLICT (key) DO NOTHING"),
[self.key(KEY_TERM_INDEX), "{}"])
row = con.execute(self.sql("SELECT value FROM {tbl} WHERE key = %s FOR UPDATE"),
[self.key(KEY_TERM_INDEX)]).fetchone()
meta = json.loads(row[0]) if row and isinstance(row[0], str) else (row[0] if row else {})
if not prepare and meta.get("version") != 2:
raise TermsNotReady(TERMS_NOT_READY)
if prepare:
self.ensure_schema(con)
yield con, meta
meta["activeTotal"] = con.execute(self.term_sql(
"SELECT COUNT(*) FROM {terms} WHERE tenant = %s AND active = TRUE"), [self.slug]).fetchone()[0]
meta["version"] = 2
self._put_meta(con, meta)
def _insert(self, con, rows):
statement = self.term_sql("INSERT INTO {terms} (tenant, term, label, active, starred) "
"SELECT * FROM unnest(%s::text[], %s::text[], %s::text[], %s::boolean[], %s::boolean[]) "
"ON CONFLICT (tenant, term) DO NOTHING")
for start in range(0, len(rows), 1000):
con.execute(statement, [list(values) for values in zip(*rows[start:start + 1000])])
def seed(self, found, legacy, *, ours=None, tracking_error=None):
with self._locked(prepare=True) as (con, meta):
if meta.get("version") != 2:
# Existing per-term rows override the older collection, including removals and stars.
cursor = ""
while True:
rows = con.execute(self.sql("SELECT key, value FROM {tbl} WHERE starts_with(key, %s) "
'AND key COLLATE "C" > %s ORDER BY key COLLATE "C" LIMIT 500'),
[self.prefix, cursor]).fetchall()
if not rows:
break
migrated = []
for key, value in rows:
value = json.loads(value) if isinstance(value, str) else value
term = _norm(value.get("term") or key[len(self.prefix):])
if term:
migrated.append((self.slug, term, display_term(term), value.get("active", True) is True,
value.get("starred", False) is True))
self._insert(con, migrated)
cursor = rows[-1][0]
removed = {_norm(t) for t in legacy.get("removed") or ()}
terms = {_norm(t) for t in [*(legacy.get("seeded") or ()),
*(legacy.get("added") or ()), *removed] if _norm(t)}
self._insert(con, [(self.slug, t, display_term(t), t not in removed, False) for t in terms])
self._insert(con, [(self.slug, t, display_term(t), True, False) for t in found if t])
# Only a checked ownership result may replace membership, including a valid empty set.
if ours is not None:
self._insert(con, [(self.slug, t, display_term(t), True, False) for t in ours])
con.execute(self.term_sql("UPDATE {terms} SET ours = (term = ANY(%s)) WHERE tenant = %s "
"AND ours IS DISTINCT FROM (term = ANY(%s))"), [list(ours), self.slug, list(ours)])
if tracking_error:
meta.update(trackingReady=False, trackingCause=tracking_error)
elif ours is not None:
meta.update(trackingReady=True, trackingCause=None)
def edit(self, terms, *, active=None, starred=None):
with self._locked() as (con, _meta):
self._insert(con, [(self.slug, t, display_term(t), True, False) for t in terms])
field, value = ("active", active) if active is not None else ("starred", starred)
with con.cursor() as cur:
cur.executemany(self.term_sql(f"UPDATE {{terms}} SET {field} = %s WHERE tenant = %s AND term = %s"),
[(value, self.slug, t) for t in terms])
def page(self, search, offset, limit, selected, *, tracked=False, termOrder=None):
if not self.ready():
raise TermsNotReady(TERMS_NOT_READY)
if termOrder not in (None, "asc", "desc"):
raise ValueError("termOrder must be asc or desc.")
if tracked:
status = self.tracking_status()
if not status["trackingReady"]:
raise SourceNotReady(status["trackingCause"], page={
"terms": [], "term": None, "termTotal": None, "termOffset": 0,
"termLimit": limit, "termLabels": {}, "termFlags": {}, **status})
membership = "(ours = TRUE OR starred = TRUE)" if tracked else "active = TRUE"
order = "term" if termOrder is None else f"label {termOrder.upper()}, term {termOrder.upper()}"
# Default pages read the transactional total and the active-term B-tree directly.
total = (self.term_sql("SELECT count(*) FROM {terms} WHERE tenant = %s AND " + membership +
(" AND strpos(lower(label), %s) > 0" if search else "")) if search or tracked else
self.sql("SELECT (value ->> 'activeTotal')::bigint FROM {tbl} WHERE key = %s"))
total_params = ([self.slug, *([display_term(search)] if search else [])]
if search or tracked else [self.key(KEY_TERM_INDEX)])
tail = ("), "
"bounds AS (SELECT total, CASE WHEN %s >= total THEN "
"greatest(0, floor((total - 1)::numeric / %s)::bigint * %s) ELSE %s END AS skip FROM totals) "
"SELECT total, skip, EXISTS(SELECT 1 FROM {terms} WHERE tenant = %s AND term = %s AND " + membership + "), "
'ARRAY(SELECT term FROM {terms} WHERE tenant = %s AND ' + membership + ' '
+ ("AND strpos(lower(label), %s) > 0 " if search else "") +
'ORDER BY ' + order + ' LIMIT %s OFFSET (SELECT skip FROM bounds)) '
"FROM bounds")
if isinstance(total, str):
statement = self.term_sql("WITH totals(total) AS (" + total + tail)
else:
# PostgreSQL's `self.sql` returns a psycopg Composed: compose it, never concatenate it as text.
from psycopg import sql as psql
statement = psql.SQL("WITH totals(total) AS (") + total + psql.SQL(self.term_sql(tail))
params = [*total_params, offset, limit, limit, offset, self.slug, selected, self.slug,
*([display_term(search)] if search else []), limit]
with self.connection() as con:
row = con.execute(statement, params).fetchone()
total, start, held, terms = row
picked = selected if held else next(iter(terms), None)
flags = self.flags([*terms, *([picked] if picked else [])])
return {"terms": list(terms), "term": selected if held else next(iter(terms), None),
"termTotal": total, "termOffset": start, "termLimit": limit,
"termLabels": {t: display_term(t) for t in flags}, "termFlags": flags,
**self.tracking_status()}
def tracking_status(self):
"""Read ownership readiness separately from the wider retention list's migration checkpoint."""
with self.connection() as con:
row = con.execute(self.sql("SELECT value ->> 'trackingReady', value ->> 'trackingCause' "
"FROM {tbl} WHERE key = %s"), [self.key(KEY_TERM_INDEX)]).fetchone()
ready = bool(row and row[0] == "true")
cause = (row[1] if row else None) or TRACKING_NOT_READY
return {"trackingReady": ready, "trackingCause": None if ready else cause}
def require_tracking(self):
status = self.tracking_status()
if not status["trackingReady"]:
raise SourceNotReady(status["trackingCause"])
def flags(self, terms):
if not terms:
return {}
with self.connection() as con:
rows = con.execute(self.term_sql("SELECT term, ours, starred, active FROM {terms} "
"WHERE tenant = %s AND term = ANY(%s)"), [self.slug, list(terms)]).fetchall()
return {term: {"ours": ours, "starred": starred, "active": active} for term, ours, starred, active in rows}
def count(self, *, tracked=False):
if not self.ready():
raise TermsNotReady(TERMS_NOT_READY)
if tracked:
self.require_tracking()
with self.connection() as con:
return con.execute(self.term_sql("SELECT COUNT(*) FROM {terms} WHERE tenant = %s AND " +
("(ours = TRUE OR starred = TRUE)" if tracked else "active = TRUE")), [self.slug]).fetchone()[0]
def all(self, *, tracked=False):
if not self.ready():
raise TermsNotReady(TERMS_NOT_READY)
with self.connection() as con:
return [r[0] for r in con.execute(self.term_sql('SELECT term FROM {terms} WHERE tenant = %s AND '
+ ('(ours = TRUE OR starred = TRUE)' if tracked else 'active = TRUE') + ' ORDER BY term'),
[self.slug]).fetchall()]
def _term_index(tenant_key, st):
from core.store_pg import PgStore
bound = _bound_store(tenant_key, st)
if getattr(st, "key", tenant_key) != tenant_key or getattr(st, "store_namespace", ""):
raise RuntimeError("The store handle belongs to another tenant.")
if isinstance(bound, TermStoreFixture):
return bound.term_store
return _TermIndex(tenant_key, st) if isinstance(bound, PgStore) else None
def _term_sql(column):
return f"lower(trim(regexp_replace(coalesce({column}, ''), '\\s+', ' ', 'g')))"
# --------------------------------------------------------------------------- mirror reads
def _mirror_path(tenant_key, st):
from harness import amazon_store
rt = st if hasattr(st, "datastore_path") else None
return amazon_store._tenant_mirror_path(tenant_key, rt)
class _ReadFailed(Exception):
"""A mirror read failed, was locked or busy. Never read as an absent table."""
class SourceNotReady(_ReadFailed):
"""A required source cannot support a checked weekly result."""
def __init__(self, message, *, page=None):
super().__init__(message)
self.page = page
@contextlib.contextmanager
def _reader(tenant_key, st):
"""A read-only cursor on the tenant mirror, None when the mirror file does not exist.
An open failure raises `_ReadFailed` naming its error class."""
path = Path(_mirror_path(tenant_key, st))
if not path.exists():
yield None
return
from harness import datastore
try:
shared = Path(datastore.DB_PATH).resolve() == path.resolve()
except Exception: # noqa: BLE001
shared = False
# The process holds one writer on its own mirror: a separate read-only connect would refuse.
try:
if shared:
cur = datastore.ro_cursor()
else:
import duckdb
cur = duckdb.connect(str(path), read_only=True)
except Exception as exc: # noqa: BLE001
raise _ReadFailed(type(exc).__name__) from exc
try:
yield cur
finally:
cur.close()
def _columns(cur, table):
"""The table's column names, an empty set when it is absent. A failed read raises `_ReadFailed`."""
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 _ReadFailed(type(exc).__name__) from exc
def _ba_columns(cur):
"""Check weekly readiness with metadata and scalar existence checks, without loading source rows."""
return _ba_source(cur)[0]
def _ba_source(cur):
"""The Brand Analytics columns and whether a dated weekly report exists. Unready sources raise."""
cols = _columns(cur, BA_TABLE) if cur is not None else set()
weekly = False
if cols and not {"_period_start", "_period_end"} <= cols:
try:
populated = cur.execute(f"SELECT EXISTS(SELECT 1 FROM {BA_TABLE})").fetchone()[0]
except Exception as exc: # noqa: BLE001
raise _ReadFailed(type(exc).__name__) from exc
if populated:
raise SourceNotReady(BA_DATES_MISSING)
elif {"_period_start", "_period_end"} <= cols:
try:
populated, weekly = cur.execute(
f"SELECT EXISTS(SELECT 1 FROM {BA_TABLE}), "
f"EXISTS(SELECT 1 FROM {BA_TABLE} WHERE {_weekly_period_sql()})").fetchone()
except Exception as exc: # noqa: BLE001
raise _ReadFailed(type(exc).__name__) from exc
if populated and not weekly:
raise SourceNotReady(BA_WEEKLY_MISSING)
return cols, bool(weekly)
def _sqp(cur, term=None):
"""Monthly SQP rows as (term, asin, impressions, end_date) for our ASINs."""
need = {"search_query_data_search_query", "asin"}
cols = _columns(cur, SQP_TABLE) if cur is not None else set()
if not need <= cols:
return []
if "_report_period" in cols:
period = '"_report_period" = \'MONTH\''
elif {"start_date", "end_date"} <= cols:
start = 'TRY_CAST("start_date" AS DATE)'
end_date = 'TRY_CAST("end_date" AS DATE)'
period = f'day({start}) = 1 AND {end_date} = last_day({start})'
else:
return []
imp = ('TRY_CAST("impression_data_asin_impression_count" AS DOUBLE)'
if "impression_data_asin_impression_count" in cols else "0")
end = 'CAST("end_date" AS VARCHAR)' if "end_date" in cols else "''"
where = f" WHERE {period}"
if term is not None:
where += f' AND {_term_sql("search_query_data_search_query")} = ?'
rows = cur.execute(
f'SELECT "search_query_data_search_query", "asin", {imp}, {end} FROM {SQP_TABLE}{where}',
[term] if term is not None else []).fetchall()
return [(_norm(t), str(a or "").strip(), float(i or 0), str(e or "")) for t, a, i, e in rows if _norm(t)]
def _our_asins(cur, sqp_rows, candidates=None):
ours = {a for _t, a, _i, _e in sqp_rows if a}
if cur is not None and candidates != set() and "asin" in _columns(cur, SQP_TABLE):
wanted = sorted(candidates) if candidates is not None else None
where = ' AND trim("asin") IN (' + ','.join('?' for _ in wanted) + ')' if wanted is not None else ''
ours |= {str(r[0]).strip() for r in cur.execute(
f'SELECT DISTINCT "asin" FROM {SQP_TABLE} WHERE "asin" IS NOT NULL{where}', wanted or []).fetchall()}
if cur is not None and candidates != set() and "asin1" in _columns(cur, LISTINGS_TABLE):
wanted = sorted(candidates) if candidates is not None else None
where = ' AND trim("asin1") IN (' + ','.join('?' for _ in wanted) + ')' if wanted is not None else ''
ours |= {str(r[0]).strip() for r in cur.execute(
f'SELECT DISTINCT "asin1" FROM {LISTINGS_TABLE} WHERE "asin1" IS NOT NULL{where}',
wanted or []).fetchall()}
return {a for a in ours if a}
def _our_asin_for(term, sqp_rows):
"""R23: our ASIN with the most impressions on the term, in the term's newest SQP period."""
rows = [r for r in sqp_rows if r[0] == term and r[1]]
if not rows:
return None
newest = max(r[3] for r in rows)
totals = {}
for _t, asin, imp, end in rows:
if end == newest:
totals[asin] = totals.get(asin, 0.0) + imp
return max(sorted(totals), key=lambda a: totals[a])
def _top3(cur, terms=None):
"""{term: {period_start: {"end": str, "rows": {asin: (rank, click_share, conversion_share)}}}}."""
cols = _ba_columns(cur)
if not {"search_term", "clicked_asin", "click_share_rank", "_period_start", "_period_end"} <= cols:
return {}
cs = 'TRY_CAST("click_share" AS DOUBLE)' if "click_share" in cols else "NULL"
cv = 'TRY_CAST("conversion_share" AS DOUBLE)' if "conversion_share" in cols else "NULL"
pe = 'CAST(TRY_CAST("_period_end" AS DATE) AS VARCHAR)'
wanted = None if terms is None else sorted({_norm(t) for t in terms})
if wanted == []:
return {}
where = (f' AND {_term_sql("search_term")} IN (SELECT unnest(?::VARCHAR[]))'
if wanted is not None else '')
rows = cur.execute(
f'SELECT "search_term", CAST(TRY_CAST("_period_start" AS DATE) AS VARCHAR), {pe}, "clicked_asin", '
f'TRY_CAST("click_share_rank" AS INTEGER), {cs}, {cv} FROM {BA_TABLE} '
f'WHERE TRY_CAST("click_share_rank" AS INTEGER) BETWEEN 1 AND 3 '
f'AND ({_weekly_period_sql()}){where}', [wanted] if wanted is not None else []).fetchall()
return _ranked_history(rows, wanted)
def _ranked_history(rows, wanted=None):
out = {}
for term, start, end, asin, rank, share, conv in rows:
key = _norm(term)
if wanted is not None and key not in wanted or not asin:
continue
week = out.setdefault(key, {}).setdefault(str(start), {"end": str(end or ""), "rows": {}})
week["rows"][str(asin).strip()] = (int(rank), share, conv)
return out
def _page_history(cur, terms, selected):
"""Read page week counts and the selected term's ranks in one source scan."""
cols = _ba_columns(cur)
wanted = sorted(set(terms) | ({selected} if selected else set()))
if not wanted or not {"search_term", "_period_start", "_period_end"} <= cols:
return {}, {}
optional = {name: f'TRY_CAST("{name}" AS {kind})' if name in cols else f'NULL::{kind}'
for name, kind in (("clicked_asin", "VARCHAR"), ("click_share_rank", "INTEGER"),
("click_share", "DOUBLE"), ("conversion_share", "DOUBLE"))}
query = (f'WITH held AS MATERIALIZED (SELECT {_term_sql("search_term")} AS term, '
'CAST(TRY_CAST("_period_start" AS DATE) AS VARCHAR) AS start, '
'CAST(TRY_CAST("_period_end" AS DATE) AS VARCHAR) AS stop, '
+ ', '.join(f'{expr} AS "{name}"' for name, expr in optional.items()) +
f' FROM {BA_TABLE} WHERE {_weekly_period_sql()} '
f'AND {_term_sql("search_term")} IN (SELECT unnest(?::VARCHAR[]))) '
'SELECT term, COUNT(DISTINCT start) AS weeks, NULL AS start, NULL AS stop, '
'NULL AS asin, NULL::INTEGER AS rank, NULL::DOUBLE AS share, NULL::DOUBLE AS conversion '
'FROM held GROUP BY term UNION ALL '
'SELECT term, NULL, start, stop, clicked_asin, click_share_rank, click_share, conversion_share '
'FROM held WHERE term = ? AND click_share_rank BETWEEN 1 AND 3')
rows = cur.execute(query, [wanted, selected]).fetchall()
counts = {term: weeks for term, weeks, *_rest in rows if weeks is not None}
history = _ranked_history([(term, *values) for term, weeks, *values in rows if weeks is None])
return counts, history.get(selected, {})
def term_page_history(tenant_key, page, *, st=None):
"""Share the page's weekly scan with selected-term detail without preparing stored state."""
st = _rt(tenant_key, st)
selected = page["term"]
if not page["terms"] and not selected:
return {}, {}
try:
with _reader(tenant_key, st) as cur:
counts, history = _page_history(cur, page["terms"], selected)
detail = term_detail(tenant_key, selected, st=st, reader=cur, history=history) if selected else {}
except Exception as exc:
cause = str(exc) if isinstance(exc, _ReadFailed) else type(exc).__name__
raise SourceNotReady(cause, page=page) from exc
return counts, detail
def _periods(cur):
if not {"_period_start", "_period_end"} <= _ba_columns(cur):
return []
rows = cur.execute(f'SELECT DISTINCT CAST(TRY_CAST("_period_start" AS DATE) AS VARCHAR) FROM {BA_TABLE} '
f'WHERE {_weekly_period_sql()}').fetchall()
return sorted((str(r[0]) for r in rows if r[0] is not None), reverse=True)
# --------------------------------------------------------------------------- watched terms
def watched_terms(tenant_key, *, st=None):
"""Seeded SQP terms plus terms a person added, minus terms a person removed (store only)."""
st = _rt(tenant_key, st)
index = _term_index(tenant_key, st)
if index is not None and index.ready():
return index.all()
if index is not None:
legacy = index.legacy_all()
if legacy is not None:
return legacy
doc = _doc(st, KEY_TERMS)
removed = {_norm(t) for t in doc.get("removed") or ()}
terms = {_norm(t) for t in [*(doc.get("seeded") or ()), *(doc.get("added") or ())]}
return sorted(t for t in terms if t and t not in removed)
def tracked_terms(tenant_key, *, st=None):
"""Read last known newest-week membership plus current stars, even during an ownership-source failure."""
st = _rt(tenant_key, st)
index = _term_index(tenant_key, st)
if index is None:
raise TermsNotReady(TERMS_NOT_READY)
return index.all(tracked=True)
def tracked_count(tenant_key, *, st=None):
"""Count the tracked subset in SQL, raising SourceNotReady when its ownership source is unchecked."""
st = _rt(tenant_key, st)
index = _term_index(tenant_key, st)
if index is None:
raise TermsNotReady(TERMS_NOT_READY)
return index.count(tracked=True)
def _brand_source_sql(cur):
"""Share one dated BA/listings join between membership and bulk click-share readers."""
need = {"search_term", "clicked_asin", "click_share_rank", "_period_start", "_period_end"}
cols, weekly = _ba_source(cur)
if cur is None or not need <= cols or "asin1" not in _columns(cur, LISTINGS_TABLE):
raise SourceNotReady("Tracked terms need dated Brand Analytics and listing ASINs.")
if not weekly:
raise SourceNotReady(BA_WEEKLY_MISSING)
return (f"WITH weekly AS (SELECT * FROM {BA_TABLE} WHERE {_weekly_period_sql()}), "
f"own_asins AS (SELECT DISTINCT trim(asin1) AS asin FROM {LISTINGS_TABLE} "
"WHERE asin1 IS NOT NULL AND trim(asin1) <> ''), "
"own_clicks AS (SELECT b.* FROM weekly b JOIN own_asins o ON trim(b.clicked_asin) = o.asin "
"WHERE TRY_CAST(b.click_share_rank AS INTEGER) BETWEEN 1 AND 3) ")
def _tracked_terms_source(cur):
query = _brand_source_sql(cur)
return {row[0] for row in cur.execute(query +
f'SELECT DISTINCT {_term_sql("search_term")} FROM own_clicks '
'WHERE TRY_CAST(_period_start AS TIMESTAMP) = '
'(SELECT MAX(TRY_CAST(_period_start AS TIMESTAMP)) FROM weekly) '
f'AND {_term_sql("search_term")} <> \'\'').fetchall()}
def brand_share_source(cur, terms, start=None, end=None):
"""Yield bulk (stored term, week start, week end, own click share) for a caller's selected terms."""
query = _brand_source_sql(cur)
if "click_share" not in _columns(cur, BA_TABLE):
raise SourceNotReady("Own-brand click share needs Brand Analytics click shares.")
wanted = sorted({_norm(term) for term in terms})
if not wanted:
return
rows = cur.execute(query +
f'SELECT {_term_sql("search_term")} AS term, CAST(_period_start AS DATE), '
'CAST(_period_end AS DATE), SUM(TRY_CAST(click_share AS DOUBLE)) FROM own_clicks '
f'WHERE {_term_sql("search_term")} IN (SELECT unnest(?)) '
'AND (? IS NULL OR CAST(_period_start AS DATE) >= CAST(? AS DATE)) '
'AND (? IS NULL OR CAST(_period_end AS DATE) <= CAST(? AS DATE)) '
'GROUP BY 1, 2, 3 ORDER BY 1, 2, 3', [wanted, start, start, end, end])
while batch := rows.fetchmany(500):
yield from batch
def _discover_terms(reader):
"""Union the two source vocabularies without treating SQP coverage as BA membership."""
if reader is None:
raise TermsNotReady(TERMS_NOT_READY)
sources = []
if "search_query_data_search_query" in _columns(reader, SQP_TABLE):
sources.append(f'SELECT {_term_sql("search_query_data_search_query")} AS term FROM {SQP_TABLE}')
if {"search_term", "_period_start", "_period_end"} <= _columns(reader, BA_TABLE):
sources.append(f'SELECT {_term_sql("search_term")} AS term FROM {BA_TABLE} WHERE {_weekly_period_sql()}')
if not sources:
raise TermsNotReady(TERMS_NOT_READY)
return {row[0] for row in reader.execute(
'SELECT DISTINCT term FROM (' + ' UNION ALL '.join(sources) + ") WHERE term <> ''").fetchall()}
def seed_terms(tenant_key, *, st=None, cur=None):
"""Discover SQP queries and held weekly BA terms while preserving authored membership."""
st = _rt(tenant_key, st)
index = _term_index(tenant_key, st)
def failure_cause(exc):
if isinstance(exc, SourceNotReady):
return str(exc)
cause = str(exc) if isinstance(exc, _ReadFailed) else type(exc).__name__
return f"Tracked terms could not be checked ({cause}). The next Amazon sync will retry."
def source(reader):
try:
found = _discover_terms(reader)
except Exception as exc:
if index is None:
raise
return set(), None, failure_cause(exc)
try:
ours = _tracked_terms_source(reader)
except Exception as exc: # noqa: BLE001
return found, None, failure_cause(exc)
return found, ours, None
if cur is None:
try:
with _reader(tenant_key, st) as reader:
found, ours, tracking_error = source(reader)
except _ReadFailed as exc:
if index is None:
raise
found, ours, tracking_error = set(), None, failure_cause(exc)
else:
found, ours, tracking_error = source(cur)
if index is not None:
index.seed(found, _doc(st, KEY_TERMS) if not index.ready() else {}, ours=ours,
tracking_error=tracking_error)
return index.all()
def _seed(doc):
doc = doc if isinstance(doc, dict) else {}
seeded = set(doc.get("seeded") or ()) | found
doc["seeded"] = sorted(seeded)
doc.setdefault("added", [])
doc.setdefault("removed", [])
doc["seeded_at"] = _stamp(_now())
return doc
st.update(KEY_TERMS, _seed)
return watched_terms(tenant_key, st=st)
def _edit_terms(tenant_key, terms, add, st, page_only=False):
clean = sorted({_norm(t) for t in terms or () if _norm(t)})
if not clean:
raise ValueError("Name at least one search term.")
st = _rt(tenant_key, st)
index = _term_index(tenant_key, st)
if index is not None:
if not index.ready():
raise TermsNotReady(TERMS_NOT_READY)
index.edit(clean, active=add)
return clean if page_only else index.all()
if page_only:
raise TermsNotReady(TERMS_NOT_READY)
def _apply(doc):
doc = doc if isinstance(doc, dict) else {}
added, removed = set(doc.get("added") or ()), set(doc.get("removed") or ())
if add:
added |= set(clean)
removed -= set(clean)
else:
removed |= set(clean)
added -= set(clean)
doc.setdefault("seeded", [])
doc["added"], doc["removed"] = sorted(added), sorted(removed)
return doc
st.update(KEY_TERMS, _apply)
return watched_terms(tenant_key, st=st)
def add_terms(tenant_key, terms, *, st=None, page_only=False):
"""Watch these terms. A term a person removed before comes back."""
return _edit_terms(tenant_key, terms, True, st, page_only)
def remove_terms(tenant_key, terms, *, st=None, page_only=False):
"""Stop watching these terms. A later SQP seed does not bring them back."""
return _edit_terms(tenant_key, terms, False, st, page_only)
def star_term(tenant_key, term, starred, *, st=None):
"""Merge a star into one authored row without changing its key or active flag."""
key = _norm(term)
if not key or type(starred) is not bool:
raise ValueError("Name a term and a boolean starred value.")
st = _rt(tenant_key, st)
index = _term_index(tenant_key, st)
if index is None or not index.ready():
raise TermsNotReady(TERMS_NOT_READY)
index.edit([key], starred=starred)
return {"term": key, "termLabel": display_term(key), **index.flags([key])[key]}
def terms_summary(tenant_key, *, st=None):
"""Every watched term with its source, newest held period and open alert count. No cap."""
st = _rt(tenant_key, st)
doc = _doc(st, KEY_TERMS)
seeded = {_norm(t) for t in doc.get("seeded") or ()}
terms = watched_terms(tenant_key, st=st)
with _reader(tenant_key, st) as cur:
held = _top3(cur, terms)
open_counts = _alert_index(tenant_key, st).open_counts()
return {"terms": [{"term": t, "termLabel": display_term(t), "source": "seeded" if t in seeded else "added",
"newestPeriod": max(held.get(t, {}), default=None),
"openAlerts": open_counts.get(t, 0)} for t in terms]}
def term_page(tenant_key, *, st=None, search="", offset=0, limit=50, selected=None, tracked=False, termOrder=None):
"""Page watched or tracked terms without materializing the vocabulary or comparing every term."""
st = _rt(tenant_key, st)
error, = _paths(tenant_key, st, KEY_WEEKS, [("last_error",)])
problems = [str(error)] if error else []
size = max(1, min(int(limit), 100))
start = max(0, int(offset))
page_error = None
try:
index = _term_index(tenant_key, st)
if index is None:
raise TermsNotReady("Search term paging needs the tenant store migration. Ask an administrator to finish it.")
controls = {}
if tracked:
controls["tracked"] = tracked
if termOrder is not None:
controls["termOrder"] = termOrder
page = index.page(_norm(search), start, size, _norm(selected), **controls)
except TermsNotReady as exc:
page_error = exc
source_error = None
try:
with _reader(tenant_key, st) as cur:
weekly = _ba_source(cur)[1]
if weekly:
problems = [problem for problem in problems if problem not in SUPERSEDED_BY_WEEKLY]
except SourceNotReady as exc:
source_error = exc
cause = READ_FAILED.format(cause=exc)
if cause not in problems:
problems.append(cause)
except _ReadFailed as exc:
cause = READ_FAILED.format(cause=exc)
if cause not in problems:
problems.append(cause)
if page_error is not None:
raise TermsNotReady(str(page_error), problems=problems) from page_error
page = {**page, "problems": problems}
if source_error is not None:
raise SourceNotReady(str(source_error), page=page) from source_error
return page
# --------------------------------------------------------------------------- settings and owner
#: Accounts that sign in for a process, never a person who can own an alert (same list as amazon_profit).
SERVICE_ACCOUNTS = ("admin", "automation", "scheduler", "qa-runner", "leadership")
def _is_service(name, rec):
name = str(name or "").strip().lower()
return (not name or name.startswith("qa") or name in SERVICE_ACCOUNTS
or name.endswith("-admin") or bool(rec.get("platform_admin")))
def owner_account(st, tenant_key=None):
"""R31: the tenant's earliest active person Administrator by display name, as money-losers use.
A tenant with no person Administrator falls back to its `admin` account, else its first Administrator."""
try:
users = st.get("users")
except Exception: # noqa: BLE001
users = None
want = str(tenant_key or "").strip().lower()
admins = [(str(name), rec) for name, rec in (users or {}).items()
if isinstance(rec, dict) and rec.get("active", True)
and str(rec.get("role") or "").lower() == "admin"
and (not want or str(rec.get("tenant") or "royal-imports").strip().lower() == want)]
people = sorted(((n, r) for n, r in admins if not _is_service(n, r)),
key=lambda item: ("@" not in item[0], not item[1].get("password_changed_at"),
str(item[1].get("password_changed_at") or ""), item[0]))
if people:
return str(people[0][1].get("name") or people[0][0])
names = sorted(n for n, _r in admins)
return "admin" if "admin" in names or not names else names[0]
def settings(tenant_key, *, st=None):
"""Thresholds and the default owner, with defaults filled in."""
st = _rt(tenant_key, st)
doc = _doc(st, KEY_SETTINGS)
out = {k: float(doc.get(k, v)) for k, v in DEFAULTS.items()}
out["defaultOwnerSetting"] = doc.get("default_owner") or None
out["defaultOwner"] = out["defaultOwnerSetting"] or owner_account(st, tenant_key)
out["statuses"] = list(STATUSES)
out["limit"] = LIMIT
return out
def set_settings(tenant_key, *, undercut_pct=None, rank_surge_pct=None, default_owner=None, st=None):
"""Set thresholds (percent, above 0 and below 100) or the default owner ('' resets to the owner account)."""
st = _rt(tenant_key, st)
changes = {}
for name, value in (("undercut_pct", undercut_pct), ("rank_surge_pct", rank_surge_pct)):
if value is None:
continue
try:
number = float(value)
except (TypeError, ValueError):
raise ValueError("A threshold must be a number.") from None
if not 0 < number < 100:
raise ValueError("A threshold must be above 0 and below 100 percent.")
changes[name] = number
def _apply(doc):
doc = doc if isinstance(doc, dict) else {}
doc.update(changes)
if default_owner is not None:
doc["default_owner"] = str(default_owner).strip() or None
return doc
st.update(KEY_SETTINGS, _apply)
return settings(tenant_key, st=st)
# --------------------------------------------------------------------------- price and rank reads
def _daily_quotes(tenant_key, st, asins):
"""Register demand and read published tenant quotes through core's daily-job interface."""
_bound_store(tenant_key, st)
if not asins:
return {}, None
try:
from core import amazon_rivals
except ImportError:
return {}, WAIT_QUOTES
try:
amazon_rivals.want(st, sorted(asins), source="competitor_watch")
return amazon_rivals.newest_quotes(st, sorted(asins)), None
except Exception as exc: # noqa: BLE001
return {}, f"Daily price job quotes could not be read ({type(exc).__name__})."
def _quote_offers(quote):
offers = quote.get("offers")
if isinstance(offers, str):
offers = json.loads(offers)
if offers is None:
return None
if not isinstance(offers, list) or any(not isinstance(offer, dict) for offer in offers):
raise ValueError("Invalid daily offer rows.")
return offers
def _quote_metric(quote, mine, cause=None):
if not quote:
return {"price": None, "currency": None, "sellerId": None, "rank": None,
"priceCause": cause or WAIT_QUOTES, "rankCause": cause or WAIT_QUOTES}
field = "offer" if mine else "buy_box"
price = quote.get(field + "_price")
price = price if isinstance(price, (int, float)) and math.isfinite(price) else None
rank = quote.get("sales_rank")
rank = rank if isinstance(rank, int) and rank > 0 else None
try:
box = next((row for row in _quote_offers(quote) or () if row.get("buyBox")), {})
except (ValueError, TypeError):
box = {}
return {"price": price, "currency": quote.get("currency"), "sellerId": box.get("sellerId"),
"priceCause": quote.get(field + "_cause") if price is not None else
quote.get(field + "_cause") or "The daily price job returned no price.",
"rank": rank, "rankCause": quote.get("rank_cause") if rank is not None else
quote.get("rank_cause") or "The daily price job returned no sales rank."}
def _listing_asins(cur):
if cur is None or "asin1" not in _columns(cur, LISTINGS_TABLE):
raise SourceNotReady("Own ASINs need Amazon listings.")
return {str(row[0]).strip() for row in cur.execute(
f"SELECT DISTINCT asin1 FROM {LISTINGS_TABLE} WHERE asin1 IS NOT NULL AND trim(asin1) <> ''").fetchall()}
def _comparison_asin(week, ours):
"""Use our highest placed listing ASIN when SQP is unavailable."""
rows = (week or {}).get("rows", {})
return min(set(rows) & ours, key=lambda asin: (rows[asin][0], asin), default=None)
def _comparison_terms(tenant_key, st):
index = _term_index(tenant_key, st)
if index is None:
raise TermsNotReady(TERMS_NOT_READY)
index.require_tracking()
return tracked_terms(tenant_key, st=st)
# --------------------------------------------------------------------------- the weekly comparison
def _seller_id(tenant_key, st):
"""Read only the seller identity from this tenant's saved Amazon credential."""
from harness import amazon_store
_bound_store(tenant_key, st)
fields = amazon_store.credential(tenant_key, st) or {}
return str(fields.get("seller_id") or "").strip()
def refresh(tenant_key, synced_tables, *, st=None, now=None):
"""Raise one durable alert per other seller from published daily offer rows."""
if "daily_quotes" not in set(synced_tables or ()):
return {"state": "not_synced", "created": []}
st = _rt(tenant_key, st)
_bound_store(tenant_key, st)
now = now or _now()
while not migrate_alerts(tenant_key, st=st):
pass
seller_id = _seller_id(tenant_key, st)
owner = settings(tenant_key, st=st)["defaultOwner"]
index = _alert_index(tenant_key, st)
if not seller_id:
index.record([], [{"term": "", "asin": None, "trigger": TRIGGERS["own_asin_seller"],
"periodStart": "offers", "cause": "Your Amazon seller ID is unavailable. Reconnect Amazon and sync again."}],
"offers", owner, now)
return {"state": "seller_not_ready", "created": []}
created = []
with _reader(tenant_key, st) as cur:
if cur is None:
return {"state": "no_mirror", "created": []}
ours = _listing_asins(cur)
if not ours:
return {"state": "no_own_asins", "created": []}
quotes, cause = _daily_quotes(tenant_key, st, ours)
raised, checks = [], []
for asin in sorted(ours):
quote = quotes.get(asin) or {}
try:
offers = _quote_offers(quote)
offer_cause = cause or (WAIT_QUOTES if not quote else quote.get("offer_cause"))
except (ValueError, TypeError):
offers, offer_cause = None, "The daily price job returned invalid offer rows."
if offers is None:
checks.append({"term": "", "asin": asin, "trigger": TRIGGERS["own_asin_seller"],
"periodStart": "offers", "cause": offer_cause or "The daily price job returned no offer rows."})
for offer in offers or ():
seller = str(offer.get("sellerId") or "").strip()
if not seller or seller == seller_id or offer.get("mine"):
continue
price, buy_box = offer.get("price"), offer.get("buyBox")
price = price if isinstance(price, (int, float)) and math.isfinite(price) else None
price_text = f"Price {price:,.2f}." if price is not None else "Price unavailable."
box_text = "Holds the Buy Box." if buy_box is True else (
"Does not hold the Buy Box." if buy_box is False else "Buy Box status unavailable.")
raised.append({"term": "", "asin": asin, "trigger": "own_asin_seller",
"triggerKey": f"own_asin_seller:{asin}:{seller}", "periodStart": str(quote.get("quote_on") or ""),
"detail": {"sellerId": seller, "price": price, "isBuyBoxWinner": buy_box,
"currency": offer.get("currency") or quote.get("currency")},
"cause": f"Seller {seller}. {price_text} {box_text}"})
created.extend(index.record(raised, checks, "offers", owner, now))
return {"state": "refreshed", "created": created}
def on_amazon_sync(tenant_key, synced_tables):
"""Compare only a new Brand Analytics week without failing the connector sync."""
failure = None
for callback in (compare,):
try:
callback(tenant_key, synced_tables)
except Exception as exc: # noqa: BLE001
failure = (READ_FAILED.format(cause=exc) if isinstance(exc, _ReadFailed) else
f"The weekly comparison failed ({type(exc).__name__}): {exc}")
if failure:
try:
def _note(doc):
doc = doc if isinstance(doc, dict) else {}
doc["last_error"] = failure
return doc
_rt(tenant_key).update(KEY_WEEKS, _note)
except Exception: # noqa: BLE001
pass
def on_quotes(tenant, changed):
"""Consume published daily quotes and retry unresolved checks within their original week."""
st = _rt(tenant)
# A seller-offer failure must not strand a waiting weekly check.
try:
refresh(tenant, ("daily_quotes",), st=st)
finally:
compare(tenant, ("daily_quotes",), st=st, quote_asins=changed)
def _clear_error(doc):
doc = doc if isinstance(doc, dict) else {}
doc.pop("last_error", None)
return doc
def _weekly_state(doc):
"""Normalize legacy timestamp week keys before comparing with connector date keys."""
out = dict(doc)
if out.get("last_compared"):
out["last_compared"] = dt.date.fromisoformat(str(out["last_compared"])[:10]).isoformat()
metrics = {}
for key, value in (out.get("metrics") or {}).items():
day = dt.date.fromisoformat(str(key)[:10]).isoformat()
if day in metrics and metrics[day] != value:
raise SourceNotReady("Stored weekly metrics disagree for one week.")
metrics[day] = value
out["metrics"] = metrics
return out
def compare(tenant_key, synced_tables=("brand_analytics",), *, st=None, now=None, quote_asins=None):
"""Compare the newest held week with the week before and raise owned alerts. Returns a summary."""
st = _rt(tenant_key, st)
now = now or _now()
while not migrate_alerts(tenant_key, st=st):
pass
synced = set(synced_tables or ())
daily = "daily_quotes" in synced
with _reader(tenant_key, st) as cur:
if cur is None:
return {"state": "no_mirror"}
if "brand_analytics" in synced:
_ba_columns(cur)
if not daily:
seed_terms(tenant_key, st=st, cur=cur)
if "brand_analytics" not in synced and not daily:
return {"state": "seeded"}
periods = _periods(cur)
if not periods:
return {"state": "no_week"}
terms = _comparison_terms(tenant_key, st)
held = _top3(cur, terms)
ours = _listing_asins(cur)
weeks = _weekly_state(_doc(st, KEY_WEEKS))
newest = periods[0]
if daily and weeks.get("last_compared") != newest:
return {"state": "awaits_weekly_pull", "period": newest}
retry = set(weeks.get("pending_quotes") or ()) & set(quote_asins or ()) if daily else None
if daily and not retry:
return {"state": "already_compared", "period": newest}
if not daily and weeks.get("last_compared") == newest and not weeks.get("last_error"):
return {"state": "already_compared", "period": newest}
previous = periods[1] if len(periods) > 1 else None
conf = settings(tenant_key, st=st)
want = {asin for term in terms for asin in ((held.get(term) or {}).get(newest) or {}).get("rows", {})}
quotes, quote_cause = _daily_quotes(tenant_key, st, want)
current = (weeks.get("metrics") or {}).get(newest) or {}
metrics = {a: current[a] if daily and a not in retry and a in current else
_quote_metric(quotes.get(a), a in ours, quote_cause) for a in sorted(want)}
prior = ((weeks.get("metrics") or {}).get(previous) or {}) if previous else {}
raised, term_checks = _term_signals(terms, held, periods, ours, metrics, prior, conf, retry)
checks = []
for asin, metric in metrics.items():
parts = list(dict.fromkeys(c for c in (metric.get("priceCause"), metric.get("rankCause")) if c))
if metric.get("price") is not None:
parts.insert(0, f"Price {metric['price']:,.2f} {metric.get('currency') or ''}.".strip())
checks.append({"term": "", "asin": asin, "trigger": "Daily price", "periodStart": newest,
"detail": metric, "cause": " ".join(parts)})
checks.extend(term_checks)
created = _alert_index(tenant_key, st).record(raised, checks, newest, conf["defaultOwner"], now)
ready = previous is not None
def _weeks(doc):
doc = _weekly_state(doc if isinstance(doc, dict) else {})
doc.setdefault("metrics", {})[newest] = metrics
doc["pending_quotes"] = [asin for asin, metric in metrics.items()
if metric.get("price") is None or metric.get("rank") is None]
doc["last_compared"] = newest
doc["comparison"] = {"ready": ready, "message": None if ready else NOT_READY,
"periodStart": newest, "previousPeriodStart": previous,
"comparedAt": _stamp(now), "alertsRaised": len(created),
"checks": len(checks)}
return _clear_error(doc)
st.update(KEY_WEEKS, _weeks)
return {"state": "compared" if ready else "waiting", "period": newest, "previous": previous,
"created": created, "checks": checks}
def _term_signals(terms, held, periods, ours, metrics, prior, conf, retry):
"""Derive weekly and quote-retry signals without changing alert state."""
raised, checks = [], []
if len(periods) < 2:
return raised, checks
newest, previous = periods[:2]
daily = retry is not None
undercut, surge = conf["undercut_pct"] / 100.0, conf["rank_surge_pct"] / 100.0
for term in terms:
now_week = (held.get(term) or {}).get(newest)
was_week = (held.get(term) or {}).get(previous)
if not now_week or not was_week:
continue
top_now, top_was = now_week["rows"], was_week["rows"]
def found(asin, key, detail, cause=None):
raised.append({"term": term, "asin": asin, "trigger": key, "periodStart": newest,
"periodEnd": now_week["end"], "previousPeriodStart": previous,
"detail": detail, "cause": cause})
def asin_cause(asin):
m = metrics.get(asin) or {}
return " ".join(c for c in (m.get("priceCause"), m.get("rankCause")) if c) or None
for asin in sorted(set(top_now) - set(top_was)) if not daily else ():
found(asin, "new_asin", {"rank": top_now[asin][0]}, asin_cause(asin))
for asin in sorted((set(top_was) - set(top_now)) & ours) if not daily else ():
found(asin, "lost_spot", {"previousRank": top_was[asin][0]})
mine = _comparison_asin(now_week, ours)
mine_m = metrics.get(mine) or {}
if mine is None and set(top_now) - ours:
checks.append({"term": term, "asin": None, "trigger": TRIGGERS["undercut"], "periodStart": newest,
"cause": "None of our listing ASINs is in this term's top 3, so there is "
"no price of ours to compare."})
for asin in sorted(set(top_now) - ours):
m = metrics.get(asin) or {}
if mine is not None:
if mine_m.get("price") is None or m.get("price") is None:
cause = " ".join(c for c in (mine_m.get("priceCause") and f"Our price: {mine_m['priceCause']}",
m.get("priceCause")) if c)
checks.append({"term": term, "asin": asin, "trigger": TRIGGERS["undercut"],
"periodStart": newest, "cause": cause or "No price was read."})
elif m["price"] <= mine_m["price"] * (1 - undercut) and (not daily or asin in retry or mine in retry):
found(asin, "undercut", {"ourAsin": mine, "ourPrice": mine_m["price"], "price": m["price"],
"percentUnder": round(100 * (1 - m["price"] / mine_m["price"]), 2)}, asin_cause(asin))
old = prior.get(asin) or {}
if m.get("rank") is None:
checks.append({"term": term, "asin": asin, "trigger": TRIGGERS["rank_surge"],
"periodStart": newest, "cause": m.get("rankCause") or "No sales rank was read."})
elif old.get("rank") is None:
if old.get("rankCause"):
checks.append({"term": term, "asin": asin, "trigger": TRIGGERS["rank_surge"],
"periodStart": newest, "cause": f"The week before: {old['rankCause']}"})
elif m["rank"] <= old["rank"] * (1 - surge) and (not daily or asin in retry):
found(asin, "rank_surge", {"rank": m["rank"], "previousRank": old["rank"]}, asin_cause(asin))
return raised, checks
# --------------------------------------------------------------------------- alerts
class _AlertIndex(_DocumentIndex):
"""Separate alert and check rows share the tenant store and one transactional sequence lock."""
@staticmethod
def decode(value):
return json.loads(value) if isinstance(value, str) else value
def ready(self):
with self.connection() as con:
row = con.execute(self.sql("SELECT value ->> 'version' FROM {tbl} WHERE key = %s"),
[self.key(KEY_ALERT_INDEX)]).fetchone()
return bool(row and row[0] == "1")
def open_counts(self):
if not self.ready():
raise AlertsNotReady(ALERTS_NOT_READY)
with self.connection() as con:
return dict(con.execute(self.sql("SELECT value ->> 'term', count(*) FROM {tbl} "
"WHERE starts_with(key, %s) AND (value ->> 'status') = 'Open' GROUP BY value ->> 'term'"),
[self.key(ALERT_PREFIX)]).fetchall())
def _put(self, con, key, value, *, keep=False):
conflict = "DO NOTHING" if keep else "DO UPDATE SET value = excluded.value"
con.execute(self.sql("INSERT INTO {tbl} (key, value) VALUES (%s, %s::jsonb) "
"ON CONFLICT (key) " + conflict), [self.key(key), json.dumps(value)])
@contextlib.contextmanager
def _locked(self, *, prepare=False):
self.pg.check_write(self.slug, "write competitor watch alert rows")
with self.connection() as con:
con.execute("SET LOCAL lock_timeout = '10s'")
con.execute("SET LOCAL statement_timeout = '30s'")
if prepare:
self._put(con, KEY_ALERT_INDEX, {"seq": 0, "itemCursor": "", "checkOffset": 0}, keep=True)
row = con.execute(self.sql("SELECT value FROM {tbl} WHERE key = %s FOR UPDATE"),
[self.key(KEY_ALERT_INDEX)]).fetchone()
meta = self.decode(row[0]) if row else {}
if not prepare and meta.get("version") != 1:
raise AlertsNotReady(ALERTS_NOT_READY)
yield con, meta
self._put(con, KEY_ALERT_INDEX, meta)
def _legacy_items(self, con, cursor, size):
return con.execute(self.sql("SELECT item.key, item.value FROM {tbl}, "
"LATERAL jsonb_each(value -> 'items') AS item WHERE {tbl}.key = %s "
"AND item.key COLLATE \"C\" > %s ORDER BY item.key COLLATE \"C\" LIMIT %s"),
[self.key(KEY_ALERTS), cursor, size]).fetchall()
def _legacy_checks(self, con, offset, size):
return con.execute(self.sql("SELECT item.ordinality, item.value FROM {tbl}, "
"LATERAL jsonb_array_elements(value -> 'checks') WITH ORDINALITY AS item(value, ordinality) "
"WHERE key = %s AND item.ordinality > %s ORDER BY item.ordinality LIMIT %s"),
[self.key(KEY_ALERTS), offset, size]).fetchall()
def migrate(self, size):
with self._locked(prepare=True) as (con, meta):
if meta.get("version") == 1:
return True
old_seq = con.execute(self.sql("SELECT value ->> 'seq' FROM {tbl} WHERE key = %s"),
[self.key(KEY_ALERTS)]).fetchone()
meta["seq"] = max(meta["seq"], int(old_seq[0] or 0) if old_seq else 0)
items = self._legacy_items(con, meta["itemCursor"], size)
for key, value in items:
alert = self.decode(value)
sequence = int(str(alert.get("id") or key)[1:])
# Missing createdAt stays missing. Copy every other legacy field unchanged.
self._put(con, ALERT_PREFIX + key, {**alert, "sequence": sequence}, keep=True)
meta["seq"] = max(meta["seq"], sequence)
meta["itemCursor"] = key
checks = self._legacy_checks(con, meta["checkOffset"], size)
for ordinal, value in checks:
self._put(con, CHECK_PREFIX + f"legacy/{ordinal}",
{**self.decode(value), "sequence": ordinal}, keep=True)
meta["checkOffset"] = ordinal
if len(items) < size and len(checks) < size:
meta["version"] = 1
return meta.get("version") == 1
def _page(self, con, prefix, search, status, offset, limit, *, checks=False,
trigger=None, include_sellers=True):
order = "period DESC, sequence, key COLLATE \"C\"" if checks else "sequence DESC"
final_order = "page.period DESC, page.sequence, page.key COLLATE \"C\"" if checks else "page.sequence DESC"
# Pick page keys before projecting alert bodies and their histories.
statement = self.sql("WITH matches AS (SELECT key, value ->> 'periodStart' AS period, "
"coalesce(value ->> 'sequence', '0')::numeric AS sequence FROM {tbl} WHERE starts_with(key, %s) "
"AND (%s::text IS NULL OR (value ->> 'status') = %s) "
"AND (%s::text IS NULL OR (value ->> 'trigger') = %s) "
"AND (%s OR coalesce(value ->> 'trigger', '') <> %s) "
"AND strpos(lower(concat_ws(' ', value ->> 'term', value ->> 'asin', value ->> 'trigger', "
"value ->> 'owner', value ->> 'cause')), %s) > 0), "
"totals AS (SELECT count(*) AS total FROM matches), bounds AS (SELECT total, "
"CASE WHEN %s >= total THEN greatest(0, floor((total - 1)::numeric / %s)::bigint * %s) "
"ELSE %s END AS skip FROM totals) SELECT total, skip, "
"ARRAY(SELECT stored.value::text FROM (SELECT key, sequence, period FROM matches ORDER BY " + order +
" LIMIT %s OFFSET (SELECT skip FROM bounds)) AS page JOIN {tbl} AS stored ON stored.key = page.key "
"ORDER BY " + final_order + ") FROM bounds")
row = con.execute(statement, [self.key(prefix), status, status, trigger, trigger, include_sellers,
TRIGGERS["own_asin_seller"], search, offset, limit, limit, offset, limit]).fetchone()
return {"total": row[0], "offset": row[1], "limit": limit,
"rows": [self.decode(value) for value in row[2]]}
def page(self, search, status, offset, limit, check_offset, check_limit, *, trigger=None, include_sellers=True):
if not self.ready():
raise AlertsNotReady(ALERTS_NOT_READY)
with self.connection() as con:
alerts = self._page(con, ALERT_PREFIX, search, status, offset, limit,
trigger=trigger, include_sellers=include_sellers)
checks = self._page(con, CHECK_PREFIX, search, None, check_offset, check_limit, checks=True,
trigger=trigger, include_sellers=include_sellers)
return {"alerts": alerts.pop("rows"), "checks": checks.pop("rows"),
"alertPage": alerts, "checkPage": checks, "statuses": list(STATUSES)}
def record(self, raised, checks, newest, owner, now):
created = []
with self._locked() as (con, meta):
for hit in raised:
seller_hit = hit["trigger"] == "own_asin_seller"
trigger_key = hit.get("triggerKey") or hit["trigger"]
latest = con.execute(self.sql("SELECT value ->> 'status' FROM {tbl} "
"WHERE starts_with(key, %s) AND (%s OR (value ->> 'term') = %s) AND (value ->> 'asin') = %s "
"AND (value ->> 'triggerKey') = %s ORDER BY (value ->> 'sequence')::numeric DESC LIMIT 1"),
[self.key(ALERT_PREFIX), seller_hit, hit["term"], hit["asin"], trigger_key]).fetchone()
if latest and (seller_hit or latest[0] in HOLDING):
continue
meta["seq"] += 1
aid = f"a{meta['seq']}"
alert = {**hit, "id": aid, "sequence": meta["seq"], "triggerKey": trigger_key,
"trigger": TRIGGERS[hit["trigger"]], "owner": owner, "status": "Open",
"due": (now.date() + dt.timedelta(days=DUE_DAYS)).isoformat(),
"createdAt": _stamp(now), "updatedAt": _stamp(now), "updatedBy": "system",
"history": [{"at": _stamp(now), "by": "system", "event": "created"}]}
self._put(con, ALERT_PREFIX + aid, alert)
created.append(aid)
con.execute(self.sql("DELETE FROM {tbl} WHERE starts_with(key, %s) AND (value ->> 'periodStart') = %s"),
[self.key(CHECK_PREFIX), newest])
for i, check in enumerate(checks):
self._put(con, CHECK_PREFIX + f"sync/{newest}/{i}", {**check, "sequence": i})
return created
def update(self, alert_id, change):
with self._locked() as (con, _meta):
row = con.execute(self.sql("SELECT value FROM {tbl} WHERE key = %s"),
[self.key(ALERT_PREFIX + alert_id)]).fetchone()
if row is None:
raise KeyError(alert_id)
alert = self.decode(row[0])
change(alert)
self._put(con, ALERT_PREFIX + alert_id, alert)
return alert
def _alert_index(tenant_key, st):
from core.store_pg import PgStore
if not isinstance(_bound_store(tenant_key, st), PgStore):
raise AlertsNotReady("Alert paging needs the tenant Postgres store. Ask an administrator to finish the store migration.")
return _AlertIndex(tenant_key, st)
def migrate_alerts(tenant_key, *, st=None, batch_size=500):
"""Copy one restartable legacy batch and return whether alert paging is ready, without schema changes."""
return _alert_index(tenant_key, _rt(tenant_key, st)).migrate(max(1, int(batch_size)))
def alerts(tenant_key, *, status=None, st=None, search="", offset=0, limit=50, check_offset=0, check_limit=50,
trigger=None, include_sellers=True):
"""Read server-filtered pages of alerts and checks without touching the legacy collection."""
if status is not None and status not in STATUSES:
raise ValueError("Unknown alert status.")
if trigger is not None and trigger not in TRIGGERS:
raise ValueError("Unknown alert trigger.")
return _alert_index(tenant_key, _rt(tenant_key, st)).page(
_norm(search), status, max(0, int(offset)), max(1, min(100, int(limit))),
max(0, int(check_offset)), max(1, min(100, int(check_limit))),
trigger=TRIGGERS.get(trigger), include_sellers=include_sellers)
def update_alert(tenant_key, alert_id, *, owner=None, status=None, due=None, user="system", st=None,
include_sellers=True):
"""Reassign, move status, or move the due date of one alert. Returns the alert."""
if status is not None and status not in STATUSES:
raise ValueError("Status must be one of: " + ", ".join(STATUSES) + ".")
if due is not None:
try:
due = dt.date.fromisoformat(str(due)).isoformat()
except ValueError:
raise ValueError("Due must be a date such as 2026-10-27.") from None
if owner is not None and not str(owner).strip():
raise ValueError("Owner cannot be empty.")
st = _rt(tenant_key, st)
def _apply(alert):
if not include_sellers and alert.get("trigger") == TRIGGERS["own_asin_seller"]:
raise PermissionError("Your account cannot change seller alerts.")
at = _stamp(_now())
event = {"at": at, "by": user}
if owner is not None:
alert["owner"] = event["owner"] = str(owner).strip()
if status is not None:
alert["status"] = event["status"] = status
if due is not None:
alert["due"] = event["due"] = due
alert["updatedAt"], alert["updatedBy"] = at, user
alert.setdefault("history", []).append(event)
return _alert_index(tenant_key, st).update(alert_id, _apply)
# --------------------------------------------------------------------------- term detail
def term_detail(tenant_key, term, *, st=None, reader=None, history=None):
"""One term's held periods, the newest top 3 with week-over-week change, and the click share trend."""
st = _rt(tenant_key, st)
key = _norm(term)
with contextlib.nullcontext(reader) if reader is not None else _reader(tenant_key, st) as cur:
sqp = _sqp(cur, key) if cur is not None else []
weeks_held = _top3(cur, [key]).get(key, {}) if history is None else history
candidates = {a for week in weeks_held.values() for a in week["rows"]}
ours = _our_asins(cur, sqp, candidates) if cur is not None else set()
periods = sorted(weeks_held, reverse=True)
metric_asins = sorted(weeks_held[periods[0]]["rows"]) if periods else []
variants = (periods[0], periods[0] + "T00:00:00Z", periods[0] + " 00:00:00") if periods else ()
state, *values = _paths(tenant_key, st, KEY_WEEKS, [("comparison",)] +
[("metrics", week, asin) for asin in metric_asins for week in variants])
metric = {asin: next((value for value in values[i * 3:i * 3 + 3] if value is not None), None)
for i, asin in enumerate(metric_asins)}
top3 = []
if periods:
newest = weeks_held[periods[0]]["rows"]
before = weeks_held[periods[1]]["rows"] if len(periods) > 1 else None
for asin, (rank, share, conv) in sorted(newest.items(), key=lambda kv: kv[1][0]):
change = None
if before is not None:
if asin not in before:
change = {"kind": "new", "previousRank": None, "clickShareDelta": None,
"conversionShareDelta": None}
else:
r0, s0, c0 = before[asin]
change = {"kind": "up" if rank < r0 else "down" if rank > r0 else "same",
"previousRank": r0,
"clickShareDelta": None if share is None or s0 is None else share - s0,
"conversionShareDelta": None if conv is None or c0 is None else conv - c0}
m = metric.get(asin) or {}
top3.append({"asin": asin, "rank": rank, "clickShare": share, "conversionShare": conv,
"isOurs": asin in ours, "change": change, "buyBoxPrice": m.get("price"),
"currency": m.get("currency"), "salesRank": m.get("rank"),
"priceCause": m.get("priceCause"), "rankCause": m.get("rankCause")})
every = sorted({a for p in periods for a in weeks_held[p]["rows"]})
trend = [{"asin": a, "isOurs": a in ours,
"clickShare": [(weeks_held[p]["rows"].get(a) or (None, None, None))[1]
if a in weeks_held[p]["rows"] else None for p in periods]} for a in every]
state = state or {}
ready = len(periods) > 1 and bool(state.get("ready", True))
return {"term": key, "termLabel": display_term(key), "periods": periods, "top3": top3, "trend": trend,
"comparison": {"ready": ready, "message": None if ready else NOT_READY}}
def _dashboard_period(start, end):
start = dt.date.fromisoformat(str(start)[:10]).isoformat() if start else None
end = dt.date.fromisoformat(str(end)[:10]).isoformat() if end else None
if start and end and start > end:
raise ValueError("The start date must come before the end date.")
return start, end
def _dashboard_term_count(tenant_key, st, *, omitted=False, independent=False):
"""Count indexed membership and read its readiness in the same tenant-bound statement."""
index = _term_index(tenant_key, st)
if index is None:
raise TermsNotReady(TERMS_NOT_READY)
membership = ("SELECT term, ours FROM {terms} WHERE tenant = %s AND (ours OR starred)"
if not independent else
"SELECT term, ours FROM {terms} WHERE tenant = %s AND ours = TRUE UNION "
"SELECT term, ours FROM {terms} WHERE tenant = %s AND starred = TRUE")
query = index.sql(index.term_sql(
"WITH tracked AS (" + membership + "), counts AS ("
"SELECT COUNT(*) AS total, COALESCE(SUM(CASE WHEN ours THEN 1 ELSE 0 END), 0) AS owned FROM tracked) "
"SELECT (SELECT value ->> 'version' FROM {tbl} WHERE key = %s), "
"(SELECT value ->> 'trackingReady' FROM {tbl} WHERE key = %s), "
"(SELECT value ->> 'trackingCause' FROM {tbl} WHERE key = %s), "
+ ("total - LEAST(10, owned)" if omitted else "total") + " FROM counts"))
with index.connection() as con:
version, ready, cause, value = con.execute(query, [index.slug] * (2 if independent else 1)
+ [index.key(KEY_TERM_INDEX)] * 3).fetchone()
if version != "2":
raise TermsNotReady(TERMS_NOT_READY)
if ready != "true":
raise SourceNotReady(cause or TRACKING_NOT_READY)
return int(value)
def _dashboard_open_alerts(tenant_key, st):
index = _alert_index(tenant_key, st)
with index.connection() as con:
rows = con.execute(index.sql(
"WITH counts AS (SELECT COALESCE(NULLIF(value ->> 'trigger', ''), 'Other') AS trigger, "
"COUNT(*) AS total FROM {tbl} WHERE starts_with(key, %s) "
"AND (value ->> 'status') = 'Open' GROUP BY 1) "
"SELECT (SELECT value ->> 'version' FROM {tbl} WHERE key = %s), trigger, total "
"FROM (SELECT 1) anchor LEFT JOIN counts ON TRUE ORDER BY trigger"),
[index.key(ALERT_PREFIX), index.key(KEY_ALERT_INDEX)]).fetchall()
if not rows or rows[0][0] != "1":
raise AlertsNotReady(ALERTS_NOT_READY)
return {trigger: int(total) for _version, trigger, total in rows if trigger is not None}
def _dashboard_click_share(cur, start, end):
"""Read the ten highest term shares and invalid evidence in one bounded BA query."""
if cur is None:
raise SourceNotReady("Own-brand click share needs dated Brand Analytics and listing ASINs.")
query = (
f"WITH newest AS (SELECT MAX(TRY_CAST(_period_start AS TIMESTAMP)) AS day FROM {BA_TABLE} "
f"WHERE {_weekly_period_sql()}), "
"held AS (SELECT b.clicked_asin, b._period_end, " + _term_sql("b.search_term") + " AS term, "
f"TRY_CAST(b.click_share AS DOUBLE) AS share FROM {BA_TABLE} b "
f"WHERE {_weekly_period_sql()} AND TRY_CAST(b._period_start AS TIMESTAMP) = (SELECT day FROM newest) "
"AND TRY_CAST(b.click_share_rank AS INTEGER) BETWEEN 1 AND 3 "
f"AND EXISTS(SELECT 1 FROM {LISTINGS_TABLE} o WHERE trim(o.asin1) = trim(b.clicked_asin)) "
"AND (? IS NULL OR CAST(b._period_start AS DATE) >= CAST(? AS DATE)) "
"AND (? IS NULL OR CAST(b._period_end AS DATE) <= CAST(? AS DATE))), "
"totals AS (SELECT term, SUM(share) AS share, "
"COUNT(*) <> COUNT(share) OR COUNT(*) <> COUNT(DISTINCT trim(clicked_asin)) "
"OR COUNT(DISTINCT _period_end) <> 1 OR term = '' "
"OR NOT BOOL_AND(isfinite(share) AND share BETWEEN 0 AND 1) "
"OR SUM(share) > 1 AS invalid FROM held GROUP BY term), "
"summary AS (SELECT COALESCE(BOOL_OR(invalid), FALSE) AS invalid FROM totals), "
"top_terms AS (SELECT term, share FROM totals ORDER BY share DESC, term LIMIT 10) "
"SELECT summary.invalid, term, share FROM summary LEFT JOIN top_terms ON TRUE "
"ORDER BY share DESC, term")
rows = cur.execute(query, [start, start, end, end]).fetchall()
if any(invalid for invalid, _term, _share in rows):
raise SourceNotReady("An own-brand click share is missing, invalid or duplicated.")
series = [{"x": display_term(term), "y": round(float(share) * 100, 9)}
for _invalid, term, share in rows if term is not None]
if not series:
raise SourceNotReady("No own-brand click share in the newest dated week.")
return series
def _check_dashboard_click_share(cur, start, end):
"""Independently rank term totals from per-ASIN weekly evidence in one source read."""
if cur is None:
raise SourceNotReady("Own-brand click share needs dated Brand Analytics and listing ASINs.")
query = (
"WITH dated AS (SELECT search_term, clicked_asin, click_share_rank, click_share, _period_start, _period_end "
f"FROM {BA_TABLE} WHERE {_weekly_period_sql()}), "
"evidence AS (SELECT " + _term_sql("search_term") + " AS term, trim(clicked_asin) AS asin, "
"_period_end AS stop, TRY_CAST(click_share AS DOUBLE) AS share FROM dated "
"WHERE TRY_CAST(_period_start AS TIMESTAMP) = (SELECT MAX(TRY_CAST(_period_start AS TIMESTAMP)) FROM dated) "
"AND TRY_CAST(click_share_rank AS INTEGER) IN (1, 2, 3) "
f"AND trim(clicked_asin) IN (SELECT DISTINCT trim(asin1) FROM {LISTINGS_TABLE} WHERE asin1 IS NOT NULL) "
"AND (? IS NULL OR CAST(_period_start AS DATE) >= CAST(? AS DATE)) "
"AND (? IS NULL OR CAST(_period_end AS DATE) <= CAST(? AS DATE))), "
"per_asin AS (SELECT term, asin, MIN(share) AS share, COUNT(*) AS n, COUNT(share) AS known, "
"MIN(stop) AS stop FROM evidence GROUP BY term, asin), "
"per_term AS (SELECT term, SUM(share) AS share, "
"MIN(n) = 1 AND MAX(n) = 1 AND MIN(known) = 1 AND term <> '' "
"AND BOOL_AND(isfinite(share) AND share >= 0 AND share <= 1) "
"AND COUNT(DISTINCT stop) = 1 AND SUM(share) <= 1 AS valid FROM per_asin GROUP BY term), "
"bad AS (SELECT COUNT(*) AS n FROM per_term WHERE valid IS NOT TRUE), "
"ranked AS (SELECT term, share, ROW_NUMBER() OVER (ORDER BY share DESC, term) AS position FROM per_term) "
"SELECT bad.n, term, share FROM bad LEFT JOIN ranked ON position <= 10 ORDER BY position")
rows = cur.execute(query, [start, start, end, end]).fetchall()
if any(bad for bad, _term, _share in rows):
raise SourceNotReady("The newest Brand Analytics share source is missing or ambiguous.")
series = [{"x": display_term(term), "y": round(100 * float(share), 9)}
for _bad, term, share in rows if term is not None]
if not series:
raise SourceNotReady("No own-brand click share in the newest dated week.")
return series
def validate_dashboard_figure(tenant_key, figure, *, start=None, end=None, st=None):
"""Reconcile each figure with separate SQL counts or period sums from its source rows."""
st = _rt(tenant_key, st)
_bound_store(tenant_key, st)
if getattr(st, "key", tenant_key) != tenant_key:
raise PermissionError("The competitor watch store belongs to another tenant.")
start, end = _dashboard_period(start, end)
fid = figure.get("id", "")
try:
if fid in ("tracked_terms", "terms_not_shown"):
value = _dashboard_term_count(tenant_key, st, omitted=fid == "terms_not_shown", independent=True)
label = "Tracked terms" if fid == "tracked_terms" else "Terms not shown"
unit, source, series = "count", TERM_TABLE, []
elif fid == "open_alerts":
index = _alert_index(tenant_key, st)
with index.connection() as con:
rows = con.execute(index.sql(
"WITH totals AS (SELECT COALESCE(NULLIF(value ->> 'trigger', ''), 'Other') AS trigger, "
"SUM(CASE WHEN value ->> 'status' = 'Open' THEN 1 ELSE 0 END) AS total FROM {tbl} "
"WHERE starts_with(key, %s) GROUP BY 1 "
"HAVING SUM(CASE WHEN value ->> 'status' = 'Open' THEN 1 ELSE 0 END) > 0) "
"SELECT (SELECT value ->> 'version' FROM {tbl} WHERE key = %s), trigger, total "
"FROM (SELECT 1) anchor LEFT JOIN totals ON TRUE ORDER BY trigger"),
[index.key(ALERT_PREFIX), index.key(KEY_ALERT_INDEX)]).fetchall()
if not rows or rows[0][0] != "1":
raise AlertsNotReady(ALERTS_NOT_READY)
series = [{"x": trigger, "y": int(count)} for _version, trigger, count in rows if trigger is not None]
value = sum(point["y"] for point in series)
label, unit, source = "Open alerts", "count", ALERT_PREFIX
elif fid == "click_share":
with _reader(tenant_key, st) as cur:
series = _check_dashboard_click_share(cur, start, end)
value = max(point["y"] for point in series)
label, unit, source = "Highest own-brand click share", "percent", BA_TABLE
else:
raise ValueError("Unknown dashboard figure.")
ok = (figure.get("value") == value and figure.get("series") == series and figure.get("label") == label
and figure.get("unit") == unit and figure.get("source_table") == source)
return {"ok": ok, "id": fid, "reason": None if ok else "The figure does not tie to its source."}
except (ValueError, RuntimeError, _ReadFailed) as exc:
return {"ok": False, "id": fid, "reason": str(exc)}
except Exception as exc:
return {"ok": False, "id": fid, "reason": f"The source check failed ({type(exc).__name__})."}
validate_figure = validate_dashboard_figure
def dashboard(tenant_key, start=None, end=None, *, st=None, **_):
"""Produce four bounded raw figures in four reads for the canonical dispatcher to check."""
st = _rt(tenant_key, st)
_bound_store(tenant_key, st)
if getattr(st, "key", tenant_key) != tenant_key:
raise PermissionError("The competitor watch store belongs to another tenant.")
start, end = _dashboard_period(start, end)
figures = []
tracked = {"id": "tracked_terms", "label": "Tracked terms", "value": None, "unit": "count",
"series": [], "source_table": TERM_TABLE}
try:
tracked["value"] = _dashboard_term_count(tenant_key, st)
except Exception as exc:
tracked["reason"] = str(exc) if isinstance(exc, (TermsNotReady, SourceNotReady)) else "Tracked terms could not be read."
figures.append(tracked)
share = {"id": "click_share", "label": "Highest own-brand click share", "value": None,
"unit": "percent", "series": [], "source_table": BA_TABLE}
try:
with _reader(tenant_key, st) as cur:
share["series"] = _dashboard_click_share(cur, start, end)
share["value"] = max(point["y"] for point in share["series"])
except Exception as exc:
share["reason"] = (str(exc) if isinstance(exc, SourceNotReady) else
"Own-brand click share needs dated Brand Analytics click shares and listing ASINs.")
figures.append(share)
omitted = {"id": "terms_not_shown", "label": "Terms not shown", "value": None,
"unit": "count", "series": [], "source_table": TERM_TABLE}
try:
omitted["value"] = _dashboard_term_count(tenant_key, st, omitted=True)
# A capacity count depends on a held series; all-starred membership does not.
if (share["value"] is None and tracked["value"] is not None
and omitted["value"] < tracked["value"]):
omitted.update(value=None, reason=share["reason"])
except Exception as exc:
omitted["reason"] = str(exc) if isinstance(exc, (TermsNotReady, SourceNotReady)) else "Tracked terms could not be read."
figures.append(omitted)
alert = {"id": "open_alerts", "label": "Open alerts", "value": None, "unit": "count",
"series": [], "source_table": ALERT_PREFIX}
try:
counts = _dashboard_open_alerts(tenant_key, st)
alert.update(value=sum(counts.values()), series=[{"x": k, "y": v} for k, v in sorted(counts.items())])
except Exception as exc:
alert["reason"] = str(exc) if isinstance(exc, AlertsNotReady) else "Open alerts could not be read."
figures.append(alert)
return {"figures": figures, "start": start, "end": end}
def validate_dashboard(tenant_key, figures=None, *, start=None, end=None, st=None, **_):
"""Expose independent figure checks to the shared dashboard dispatcher."""
st = _rt(tenant_key, st)
if figures is None:
figures = dashboard(tenant_key, start, end, st=st)["figures"]
detail = [validate_dashboard_figure(tenant_key, figure, start=start, end=end, st=st) for figure in figures]
return {"ok": bool(detail) and all(check["ok"] for check in detail), "detail": detail}
"""Competitor watch App: watched terms, weekly comparisons and owned seller alerts.
Store documents (tenant runtime store, App owned, never the mirror and never Amazon):
competitor_watch_settings {undercut_pct, rank_surge_pct, default_owner}
competitor_watch_terms {seeded: [...], added: [...], removed: [...], seeded_at}
competitor_watch_alerts legacy source, retained for explicit resumable migration
competitor_watch_alert/<id> one alert per store row
competitor_watch_check/<id> one check per store row
competitor_watch_alerts_index migration checkpoint and serialized alert sequence
competitor_watch_weeks {last_compared, comparison: {...}, metrics: {period_start: {asin: metric}}}
competitor_watch_terms_index {version: 2, activeTotal, trackingReady, trackingCause}, prepared by sync
competitor_watch_terms SQL table (tenant, term, label, active, starred, ours)
competitor_watch_term/<term> legacy rows, retained after one-time migration
Top 3 history is read from the connector table `amzn_brand_analytics`, which the store's weekly
retention keeps for watched terms. App documents never change the connector (rule 13).
Alerts and checks page in the tenant's existing Postgres store. No collection document or row cap
limits their growth. GET never prepares rows or reads the legacy collection.
"""
import contextlib
import datetime as dt
import json
import math
import re
from pathlib import Path
from urllib.parse import unquote_plus
from core.amazon_rivals import weekly_period_sql as _weekly_period_sql
TRIGGERS = {
"new_asin": "New ASIN in top 3",
"lost_spot": "Lost spot",
"undercut": "Undercut",
"rank_surge": "Rank surge",
"own_asin_seller": "Seller on my ASIN",
}
STATUSES = ("Open", "Repriced", "Ads changed", "Listing improved", "Watching", "Ignore", "Reported to Amazon")
#: A key (term, asin, trigger) is held while its latest alert has one of these statuses.
HOLDING = ("Open", "Watching", "Ignore", "Reported to Amazon")
DEFAULTS = {"undercut_pct": 5.0, "rank_surge_pct": 50.0}
DUE_DAYS = 7
NOT_READY = "Comparison starts after the next weekly pull."
#: The mirror holds the tables but this read failed or was busy. `{cause}` is the error class.
READ_FAILED = ("The Amazon mirror could not be read ({cause}), so no search term was compared "
"and the next Amazon sync will retry.")
BA_DATES_MISSING = "Brand Analytics report dates are missing"
BA_WEEKLY_MISSING = "Brand Analytics has no dated weekly reports"
#: Stored sync failures that a live probe finding a dated weekly report proves superseded.
SUPERSEDED_BY_WEEKLY = frozenset(READ_FAILED.format(cause=cause) for cause in (BA_DATES_MISSING, BA_WEEKLY_MISSING))
LIMIT = "Alerts and checks use separate store rows and server pages, with no row cap."
KEY_SETTINGS = "competitor_watch_settings"
KEY_TERMS = "competitor_watch_terms"
KEY_ALERTS = "competitor_watch_alerts"
KEY_WEEKS = "competitor_watch_weeks"
KEY_TERM_INDEX = "competitor_watch_terms_index"
TERM_PREFIX = "competitor_watch_term/"
TERM_TABLE = "competitor_watch_terms"
TERMS_NOT_READY = "Search terms are not ready. The next Amazon sync prepares term paging."
TRACKING_NOT_READY = "Tracked terms have no checked ownership source. The next Amazon sync will retry."
KEY_ALERT_INDEX = "competitor_watch_alerts_index"
ALERT_PREFIX = "competitor_watch_alert/"
CHECK_PREFIX = "competitor_watch_check/"
ALERTS_NOT_READY = "Alert paging is not ready. The next Amazon sync or an administrator must finish alert migration."
BA_TABLE = "amzn_brand_analytics"
SQP_TABLE = "amzn_search_query_performance"
LISTINGS_TABLE = "amzn_listings"
WAIT_QUOTES = "Waits for the daily price job"
def _now():
return dt.datetime.now(dt.timezone.utc)
def _stamp(at):
return at.strftime("%Y-%m-%dT%H:%M:%SZ")
def _norm(term):
return " ".join(str(term or "").split()).lower()
def display_term(term):
"""Decode a label without changing the stored parent-record key."""
return " ".join(unquote_plus(str(term or "")).split())
def _rt(tenant_key, st=None):
"""The tenant runtime store handle (get and update), or raise when it cannot be resolved."""
if st is not None:
return st
from harness import amazon_store
rt = amazon_store._runtime_for(tenant_key)
if rt is None:
raise RuntimeError(f"The store for tenant {tenant_key!r} could not be opened.")
return rt