Files
pgz-sport/workers/enrichment_worker.py
T
CC6 Worker 9c5116eaa3 M12.5 R4: coverage<70 picker + confidence>=0.7 gate + /var/log target
- Coverage computed in SQL (filled_keys * 100 / total_keys); only rows below
  threshold (default 70%, override ENRICHER_COVERAGE_MAX) are queued.
- Per-row confidence is the max of source weights (semafor.hns.family=0.95,
  wikipedia.hr=0.80, sport-pgz.hr=0.55) plus a small evidence-count bonus.
  Below threshold (default 0.70, override ENRICHER_CONFIDENCE), only 'hard'
  structured fields (profile_url, source_url, slika_url, hns_igrac_id) are
  applied — never an LLM-synthesised biografija.
- Logs now mirrored to /var/log/pgz-sport-enricher.log alongside the project
  log, so 'tail /var/log/pgz-sport-enricher.log' works as the brief asks.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-05 00:45:48 +02:00

274 lines
9.7 KiB
Python

#!/usr/bin/env python3
"""
enrichment_worker.py — 24/7 background enrichment for PGŽ Sport
Author: Damir Radulić (damir@rinet.one) / dradulic@outlook.com
Date: 2026-05-04
Version: 1.0.0
Polls pgz_sport.clanovi / klubovi / savezi for under-enriched rows, then
calls the live HTTP endpoints (POST /sport/api/v2/enrich/{kind}/{id} and
.../apply) so every row goes through the same pipeline (and audit log)
that the UI uses. This avoids forking the enrichment logic.
Selection rules (per cycle):
- sportas: clanovi rows missing profile_url AND (source IN ('hns_semafor','manual')
OR vanjski_id ? 'hns_comet') ORDER BY random() LIMIT 25
- klub: klubovi rows whose metadata.enriched_at is NULL ORDER BY random() LIMIT 10
- savez: savezi rows whose metadata.enriched_at is NULL ORDER BY random() LIMIT 5
Sleep 300 s between cycles (configurable via ENRICHER_SLEEP env).
Heartbeat to redis (cc:pgz-enricher:heartbeat) and log every cycle to
/opt/pgz-sport/_logs/enrichment_worker.log.
"""
from __future__ import annotations
import json
import os
import random
import sys
import time
import urllib.parse
import urllib.request
from datetime import datetime, timezone
import psycopg2
import psycopg2.extras
API_BASE = os.environ.get('PGZ_API_BASE', 'http://localhost:8095')
SLEEP_S = int(os.environ.get('ENRICHER_SLEEP', '300'))
DRY = os.environ.get('ENRICHER_DRY', '0') == '1'
USER_HDR = os.environ.get('ENRICHER_USER', 'enricher@pgz.local')
LOG_PATHS = [
'/var/log/pgz-sport-enricher.log',
'/opt/pgz-sport/_logs/enrichment_worker.log',
]
CONFIDENCE_MIN = float(os.environ.get('ENRICHER_CONFIDENCE', '0.7'))
COVERAGE_MAX = int(os.environ.get('ENRICHER_COVERAGE_MAX', '70'))
_pgh = os.environ.get('PG_HOST', '10.10.0.2')
_pgp = int(os.environ.get('PG_PORT', '6432'))
if _pgh in ('localhost', '127.0.0.1'):
_pgh = os.environ.get('DB_HOST', '10.10.0.2')
_pgp = int(os.environ.get('DB_PORT', '6432'))
DB = dict(host=_pgh, port=_pgp,
dbname=os.environ.get('PG_DB', 'rinet_v3'),
user=os.environ.get('PG_USER', 'rinet'),
password=os.environ.get('PG_PASS', 'R1net2026!SecureDB#v7'))
def _log(msg: str) -> None:
line = f"{datetime.now(timezone.utc).isoformat()}Z {msg}"
print(line, flush=True)
for p in LOG_PATHS:
try:
os.makedirs(os.path.dirname(p), exist_ok=True)
with open(p, 'a') as f:
f.write(line + "\n")
except Exception:
pass
def _heartbeat() -> None:
try:
import redis
r = redis.Redis(host=os.environ.get('REDIS_HOST', 'localhost'),
port=int(os.environ.get('REDIS_PORT', '6379')),
password=os.environ.get('REDIS_PASS', None))
r.set('cc:pgz-enricher:heartbeat', str(int(time.time())))
except Exception:
pass
def _db():
c = psycopg2.connect(**DB); c.autocommit = True; return c
# Coverage = (filled key fields) / (total key fields) * 100. Keep these in sync
# with enrich_router.enrich_preview() which surfaces the same scores in the UI.
_KLUB_KEYS = ('oib','sport','grad','predsjednik','tajnik','web','email','telefon',
'sjediste','godina_osnutka','ciljevi','opis_djelatnosti')
_SAVEZ_KEYS = ('oib','sport','predsjednik','tajnik','email','telefon','web',
'adresa','godina_osnutka')
# Coverage for sportas — fields the user actually wants populated.
_SPORTAS_KEYS = ('sport','profile_url','slika_url','hns_igrac_id','biografija',
'datum_rodenja','mjesto_rodenja','broj_dresa')
def _coverage_expr(table_keys: tuple[str, ...]) -> str:
"""Postgres expression that returns 0..100 coverage % for the row."""
parts = []
for k in table_keys:
parts.append(f"(CASE WHEN {k} IS NOT NULL AND ({k}::text) <> '' THEN 1 ELSE 0 END)")
total = len(table_keys)
return f"((({' + '.join(parts)})::numeric * 100) / {total})"
def _pick_sportas(limit: int = 50) -> list[int]:
"""Athletes with coverage<COVERAGE_MAX, randomly ordered."""
cov = _coverage_expr(_SPORTAS_KEYS)
sql = f"""
SELECT id FROM pgz_sport.clanovi
WHERE aktivan = TRUE
AND {cov} < %s
AND (
source IN ('hns_semafor','hns_family','manual','godisnjak')
OR jsonb_exists(vanjski_id, 'hns_comet')
OR (source_url ILIKE '%%semafor.hns.family%%')
OR profile_url ILIKE '%%semafor.hns.family%%'
)
AND ((metadata->>'enriched_at') IS NULL
OR (metadata->>'enriched_at')::timestamptz < now() - interval '7 days')
ORDER BY random()
LIMIT %s
"""
with _db() as c, c.cursor() as cur:
cur.execute(sql, (COVERAGE_MAX, limit))
return [r[0] for r in cur.fetchall()]
def _pick_klub(limit: int = 50) -> list[int]:
cov = _coverage_expr(_KLUB_KEYS)
sql = f"""
SELECT id FROM pgz_sport.klubovi
WHERE aktivan = TRUE
AND {cov} < %s
AND ((metadata->>'enriched_at') IS NULL
OR (metadata->>'enriched_at')::timestamptz < now() - interval '14 days')
ORDER BY random()
LIMIT %s
"""
with _db() as c, c.cursor() as cur:
cur.execute(sql, (COVERAGE_MAX, limit))
return [r[0] for r in cur.fetchall()]
def _pick_savez(limit: int = 50) -> list[int]:
cov = _coverage_expr(_SAVEZ_KEYS)
sql = f"""
SELECT id FROM pgz_sport.savezi
WHERE {cov} < %s
AND ((metadata->>'enriched_at') IS NULL
OR (metadata->>'enriched_at')::timestamptz < now() - interval '14 days')
ORDER BY random()
LIMIT %s
"""
with _db() as c, c.cursor() as cur:
cur.execute(sql, (COVERAGE_MAX, limit))
return [r[0] for r in cur.fetchall()]
def _http_post(path: str, body: dict | None = None) -> dict | None:
url = API_BASE.rstrip('/') + path
data = json.dumps(body or {}).encode('utf-8')
req = urllib.request.Request(
url, data=data, method='POST',
headers={'Content-Type': 'application/json',
'X-User-Email': USER_HDR})
try:
with urllib.request.urlopen(req, timeout=60) as r:
return json.loads(r.read().decode('utf-8'))
except Exception as e:
_log(f"POST {path} failed: {type(e).__name__}: {e}")
return None
# Per-source confidence weights. Anything written by an HNS Semafor /igraci/
# page is structured + verified, so we trust it implicitly. Wikipedia summaries
# are mostly safe but free-form. sport-pgz.hr "O nama" pages tend to be the
# zajednica generic info, so we down-weight them so a plain DeepSeek synthesis
# off a single sport-pgz.hr source falls below the gate.
_SOURCE_WEIGHTS = {
'semafor.hns.family': 0.95,
'wikipedia.hr': 0.80,
'sport-pgz.hr': 0.55,
}
# Fields that are safe to auto-write even from low-confidence sources because
# they come from the entity's own structured page (URLs, IDs).
_HARD_FIELDS = {'profile_url','source_url','slika_url','hns_igrac_id'}
def _confidence(proposed: dict, sources: list[dict]) -> float:
"""Crude 0..1 score: max source weight, scaled by evidence count."""
if not proposed:
return 0.0
weights = []
for s in sources or []:
w = _SOURCE_WEIGHTS.get((s.get('source') or '').lower(), 0.50)
weights.append(w)
if not weights:
return 0.0
base = max(weights)
bonus = min(0.10, 0.03 * (len(sources) - 1))
return min(1.0, base + bonus)
def _process(kind: str, eid: int) -> tuple[int, list[str]]:
"""Preview → confidence gate → apply. Returns (#applied, fields)."""
preview = _http_post(f'/api/v2/enrich/{kind}/{eid}', {})
if not preview:
return (0, [])
proposed = preview.get('proposed') or {}
sources = preview.get('sources') or []
if not proposed:
return (0, [])
conf = _confidence(proposed, sources)
# Always allow hard structured fields (URLs / IDs) — they are objective.
hard = {k: v for k, v in proposed.items() if k in _HARD_FIELDS}
soft = {k: v for k, v in proposed.items() if k not in _HARD_FIELDS}
fields = dict(hard)
if conf >= CONFIDENCE_MIN:
fields.update(soft)
if not fields:
_log(f" {kind}#{eid} skipped — confidence {conf:.2f} < {CONFIDENCE_MIN:.2f}")
return (0, [])
res = _http_post(f'/api/v2/enrich/{kind}/{eid}/apply',
{'fields': fields, 'sources': sources})
if not res or 'applied' not in res:
return (0, [])
applied = res['applied']
if applied:
_log(f" {kind}#{eid} conf={conf:.2f} → +{len(applied)} {','.join(applied.keys())}")
return (len(applied), list(applied.keys()))
def _cycle() -> dict:
started = time.time()
out = {'sportas': 0, 'klub': 0, 'savez': 0, 'fields_total': 0}
fields_total = 0
for kind, picker, limit in (
('sportas', _pick_sportas, 50),
('klub', _pick_klub, 20),
('savez', _pick_savez, 5),
):
ids = picker(limit)
random.shuffle(ids)
_log(f"cycle: {kind} candidates={len(ids)} coverage<{COVERAGE_MAX} conf>={CONFIDENCE_MIN}")
for eid in ids:
if DRY:
continue
n, fields = _process(kind, eid)
out[kind] += 1
fields_total += n
time.sleep(1.5) # gentle pacing
_heartbeat()
out['fields_total'] = fields_total
out['elapsed_s'] = round(time.time() - started, 1)
return out
def main() -> int:
_log(f"enrichment_worker starting | API_BASE={API_BASE} | sleep={SLEEP_S}s | dry={DRY}")
while True:
try:
stats = _cycle()
_log(f"cycle done: {json.dumps(stats)}")
except Exception as e:
_log(f"cycle FAILED: {type(e).__name__}: {e}")
_heartbeat()
time.sleep(SLEEP_S)
if __name__ == '__main__':
sys.exit(main() or 0)