ops: add SQL query test script and macro reclassify/reaggregate runner

- scripts/test_saved_queries.py: tests all 24 saved SQL explorer queries against the live Trino API (all 24 pass) - scripts/run_reclassify_and_reaggregate.sh: self-contained script to re-classify macro events with updated prompts and re-aggregate all tickers. Scales aggregation to 16 pods, monitors queues, scales back.
2026-04-17 08:03:14 +00:00
parent 98de4ee7c9
commit 1842b2039c
2 changed files with 154 additions and 0 deletions
@@ -0,0 +1,104 @@
+#!/bin/bash
+# Re-classify macro events and re-aggregate all tickers.
+# Run from gremlin-1 (or anywhere with kubectl access to the cluster).
+set -euo pipefail
+NS="stonks-oracle"
+
+SCHED_POD=$(kubectl get pods -n $NS -l app=scheduler -o jsonpath='{.items[0].metadata.name}')
+echo "Using scheduler pod: $SCHED_POD"
+
+echo ""
+echo "=== Step 1: Re-classify macro events ==="
+echo "Deleting old global_events + macro_impact_records, then enqueuing docs..."
+kubectl exec -n $NS "$SCHED_POD" -- python -c "
+import asyncio, json, logging
+import asyncpg, redis.asyncio as aioredis
+logging.basicConfig(level=logging.INFO, format='%(asctime)s %(levelname)s %(message)s')
+log = logging.getLogger('reclassify')
+
+async def run():
+    pool = await asyncpg.create_pool(dsn='postgresql://stonks:St0nks0racl3!@postgresql-rw.postgresql-service.svc.cluster.local:5432/stonks', min_size=1, max_size=4)
+    r = aioredis.from_url('redis://redis-master.redis-service.svc.cluster.local:6379/0')
+    ic = await pool.fetchval('SELECT count(*) FROM macro_impact_records')
+    ec = await pool.fetchval('SELECT count(*) FROM global_events')
+    dc = await pool.fetchval(\"SELECT count(*) FROM documents WHERE document_type = 'macro_event' AND status != 'rejected'\")
+    log.info('Current: %d events, %d impacts, %d macro docs', ec, ic, dc)
+    await pool.execute('DELETE FROM macro_impact_records')
+    log.info('Deleted macro_impact_records')
+    await pool.execute('DELETE FROM global_events')
+    log.info('Deleted global_events')
+    rows = await pool.fetch(\"SELECT id FROM documents WHERE document_type = 'macro_event' AND status != 'rejected' ORDER BY published_at DESC\")
+    n = 0
+    for row in rows:
+        await r.rpush('stonks:queue:macro_classification', json.dumps({'document_id': str(row['id'])}))
+        n += 1
+    log.info('Enqueued %d macro re-classification jobs', n)
+    log.info('Queue depth: %d', await r.llen('stonks:queue:macro_classification'))
+    await pool.close()
+    await r.aclose()
+
+asyncio.run(run())
+"
+
+echo ""
+echo "=== Step 2: Wait for macro classification queue to drain ==="
+echo "Extractor pods will process these. Polling queue depth..."
+while true; do
+    DEPTH=$(kubectl exec -n $NS "$SCHED_POD" -- python -c "
+import redis; r=redis.from_url('redis://redis-master.redis-service.svc.cluster.local:6379/0'); print(r.llen('stonks:queue:macro_classification'))
+" 2>/dev/null)
+    echo "  Macro queue: $DEPTH"
+    if [ "$DEPTH" = "0" ]; then
+        echo "Macro queue drained!"
+        break
+    fi
+    sleep 15
+done
+
+echo ""
+echo "=== Step 3: Scale aggregation to 16 pods ==="
+kubectl scale deployment/aggregation --replicas=16 -n $NS
+kubectl rollout status deployment/aggregation -n $NS --timeout=120s
+
+echo ""
+echo "=== Step 4: Enqueue re-aggregation for all tickers ==="
+kubectl exec -n $NS "$SCHED_POD" -- python -c "
+import asyncio, json, logging
+import asyncpg, redis.asyncio as aioredis
+logging.basicConfig(level=logging.INFO, format='%(asctime)s %(levelname)s %(message)s')
+log = logging.getLogger('reaggregate')
+
+async def run():
+    pool = await asyncpg.create_pool(dsn='postgresql://stonks:St0nks0racl3!@postgresql-rw.postgresql-service.svc.cluster.local:5432/stonks', min_size=1, max_size=4)
+    r = aioredis.from_url('redis://redis-master.redis-service.svc.cluster.local:6379/0')
+    rows = await pool.fetch('SELECT ticker FROM companies WHERE active = TRUE ORDER BY ticker')
+    n = 0
+    for row in rows:
+        await r.rpush('stonks:queue:aggregation', json.dumps({'ticker': row['ticker'], 'reaggregate': True}))
+        n += 1
+    log.info('Enqueued %d aggregation jobs', n)
+    log.info('Queue depth: %d', await r.llen('stonks:queue:aggregation'))
+    await pool.close()
+    await r.aclose()
+
+asyncio.run(run())
+"
+
+echo ""
+echo "=== Step 5: Monitor aggregation queue ==="
+while true; do
+    DEPTH=$(kubectl exec -n $NS "$SCHED_POD" -- python -c "
+import redis; r=redis.from_url('redis://redis-master.redis-service.svc.cluster.local:6379/0'); print(r.llen('stonks:queue:aggregation'))
+" 2>/dev/null)
+    echo "  Aggregation queue: $DEPTH"
+    if [ "$DEPTH" = "0" ]; then
+        echo "Aggregation queue drained!"
+        break
+    fi
+    sleep 10
+done
+
+echo ""
+echo "=== Step 6: Scale aggregation back to 4 pods ==="
+kubectl scale deployment/aggregation --replicas=4 -n $NS
+echo "Done! Re-classification and re-aggregation complete."