broad-announce/scripts/m7_smoke.sh

#!/usr/bin/env bash
# Live M7 smoke test. Run from repo root:
#   bash scripts/m7_smoke.sh
#
# Walks through the 4 scenarios in M7_VERIFICATION.md:
#
#   Step 2 — hypertable: query timescaledb_information.hypertables,
#            expect `deliveries` listed.
#   Step 3 — retention policy: query
#            timescaledb_information.jobs, expect a 7-day
#            retention policy on `deliveries`.
#   Step 4 — archiver dry-run: insert a synthetic 8d-old
#            row, run archiverd once, expect the row to
#            move from Postgres to ClickHouse and the
#            Postgres count to drop by 1.
#   Step 5 — ClickHouse archive: SELECT count(*), min, max
#            from ba_archive.deliveries_archive; expect
#            ≥1 row with created_at in the 7-30d range.
#
# Exit code is the number of failed checks.

set -e
cd "$(dirname "$0")/.."

PGCMD="docker exec -i broad-announce-postgres-1 psql -U ba -d ba -A -t"
CHURL=http://localhost:8123
CHCMD="curl -sS -X POST $CHURL/"

fails=0
pass() { echo "  ✅ $*"; }
fail() { echo "  ❌ $*"; fails=$((fails+1)); }

ch_query() {
  $CHCMD --data-binary "$1" 2>/dev/null
}

# ── Setup: build archiverd fresh (must be v4 or later) ──
ARCHIVERD_BIN=/tmp/archiverd-m7
if [[ ! -x $ARCHIVERD_BIN ]]; then
  echo "▸ Building $ARCHIVERD_BIN"
  CGO_ENABLED=0 go build -o $ARCHIVERD_BIN ./cmd/archiverd
fi
# Always copy the fresh build into the container. Earlier
# dev runs may have left a stale binary at /tmp/archiverd.
docker exec broad-announce-ingestd-1 rm -f /tmp/archiverd 2>/dev/null || true

# ─────────────────────────────────────────────────────────────────
echo "── M7 smoke — Timescale 7d hot + ClickHouse archive ──"
echo ""

# ── Step 2: hypertable exists ──────────────────────────────
echo "── Step 2: deliveries is a Timescale hypertable ──"
hypertables=$($PGCMD -c "SELECT hypertable_name FROM timescaledb_information.hypertables;" 2>/dev/null | head -5)
if echo "$hypertables" | grep -q "deliveries"; then
  pass "deliveries is a hypertable"
else
  fail "deliveries NOT in hypertables list: $hypertables"
fi

# ── Step 3: retention policy is 7d ─────────────────────────
echo ""
echo "── Step 3: 7-day retention policy is active ──"
jobs=$($PGCMD -c "SELECT application_name, config FROM timescaledb_information.jobs WHERE config::text LIKE '%7 days%';" 2>/dev/null | head -5)
if echo "$jobs" | grep -q "7 days"; then
  pass "7-day retention policy present ($jobs)"
else
  fail "7-day retention policy not found in jobs: $jobs"
fi

# ── Step 4: archiver moves a synthetic 8d-old row ─────────
echo ""
echo "── Step 4: 8d-old synthetic row moves to ClickHouse ──"
# Insert a unique row, run archiverd, verify it moved.
TAG="m7-step4-$(date +%s%N)"
$PGCMD -c "INSERT INTO deliveries (alert_id, company_id, individual_id, channel, target, status, attempts, last_error, payload, created_at, sent_at) VALUES ('$TAG', 'acme-001', 'ind-acme-001', 'fcm', 'fake-token', 'sent', 0, '', '{\"smoke_tag\":\"$TAG\"}'::jsonb, now() - INTERVAL '8 days', now() - INTERVAL '8 days') ON CONFLICT DO NOTHING;" >/dev/null 2>&1
pg_before=$($PGCMD -c "SELECT count(*) FROM deliveries WHERE alert_id = '$TAG';" 2>/dev/null | head -1)
ch_before=$(ch_query "SELECT count(*) FROM ba_archive.deliveries_archive WHERE alert_id = '$TAG' FORMAT TabSeparated" 2>/dev/null | head -1)
if [[ "$pg_before" == "1" ]]; then
  pass "synthetic row in postgres (count=1)"
else
  fail "synthetic row not in postgres: count=$pg_before"
fi

# Run archiverd once. We use --port 8807 to avoid clashing
# with anything left running in dev.
docker exec broad-announce-ingestd-1 pkill -9 -f archiverd 2>/dev/null || true
sleep 1
docker cp $ARCHIVERD_BIN broad-announce-ingestd-1:/tmp/archiverd 2>/dev/null
timeout 10 docker exec \
  -e BA_ENV=dev -e BA_HTTP_ADDR=:8807 \
  -e BA_POSTGRES_DSN="postgres://ba:ba@postgres:5432/ba?sslmode=disable" \
  -e BA_ARCHIVERD_CLICKHOUSE_URL="http://clickhouse:8123" \
  -e BA_ARCHIVERD_RUN_EVERY_SECONDS=999999 \
  -e BA_ARCHIVERD_OLDER_THAN_HOURS=168 \
  -e BA_ARCHIVERD_BATCH_SIZE=10000 \
  -e BA_LOG_LEVEL=info \
  broad-announce-ingestd-1 /tmp/archiverd > /tmp/m7-archiverd.log 2>&1 || true
if grep -q "archiver run ok" /tmp/m7-archiverd.log; then
  pass "archiverd run completed ok"
else
  fail "archiverd did not complete; log tail:"
  tail -10 /tmp/m7-archiverd.log | sed 's/^/      /'
fi

pg_after=$($PGCMD -c "SELECT count(*) FROM deliveries WHERE alert_id = '$TAG';" 2>/dev/null | head -1)
ch_after=$(ch_query "SELECT count(*) FROM ba_archive.deliveries_archive WHERE alert_id = '$TAG' FORMAT TabSeparated" 2>/dev/null | head -1)
if [[ "$pg_after" == "0" ]]; then
  pass "row moved out of postgres (count=0 after)"
else
  fail "row still in postgres: count=$pg_after"
fi
if [[ "$ch_after" == "1" ]]; then
  pass "row landed in ClickHouse (count=1)"
else
  fail "row missing from ClickHouse: count=$ch_after (was $ch_before)"
fi

# ── Step 5: ClickHouse aggregate query ─────────────────────
echo ""
echo "── Step 5: ClickHouse archive queryable ──"
archive_count=$(ch_query "SELECT count(*) FROM ba_archive.deliveries_archive FORMAT TabSeparated" 2>/dev/null | head -1)
if [[ ${archive_count:-0} -ge 1 ]]; then
  pass "ba_archive.deliveries_archive count = $archive_count (≥1 row archived)"
else
  fail "ClickHouse archive empty (count=$archive_count)"
fi
min_age=$(ch_query "SELECT dateDiff('day', min(created_at), now()) FROM ba_archive.deliveries_archive FORMAT TabSeparated" 2>/dev/null | head -1)
if [[ ${min_age:-999} -ge 7 ]]; then
  pass "oldest archived row is $min_age days old (≥7)"
else
  fail "oldest archived row is only $min_age days old (expected ≥7)"
fi

# ── Summary ─────────────────────────────────────────────────
echo ""
if [[ $fails -eq 0 ]]; then
  echo "🟢 M7 smoke PASS — all checks green"
  exit 0
else
  echo "🔴 M7 smoke FAIL — $fails check(s) failed"
  exit $fails
fi