lore-engine-poc/scripts/ci-smoke.sh

#!/usr/bin/env bash
# lore-engine-poc — CI smoke runner
#
# Brings up the stack from a clean working tree, waits for all four services
# to be healthy, runs the seed, runs test.sh, and exits 0/1.
#
# Designed to be run identically on a developer laptop, in CI, or in a
# one-off cron. See docs/SMOKE.md for the full rationale + troubleshooting.
#
# Usage:
#   ./scripts/ci-smoke.sh                 # full bring-up + test + teardown
#   ./scripts/ci-smoke.sh --keep-up       # leave the stack running on success
#   ./scripts/ci-smoke.sh --skip-build    # skip `docker compose build`
#
# Exit codes:
#   0  smoke passed
#   1  a service did not become healthy in time
#   2  seed.py failed
#   3  test.sh failed
#   4  usage / argument error
#   5  docker compose not available

set -euo pipefail

# ─── argument parsing ────────────────────────────────────────────────────────
KEEP_UP=0
SKIP_BUILD=0
for arg in "$@"; do
  case "$arg" in
    --keep-up)    KEEP_UP=1 ;;
    --skip-build) SKIP_BUILD=1 ;;
    -h|--help)
      sed -n '2,18p' "$0" | sed 's/^# \{0,1\}//'
      exit 0
      ;;
    *)
      echo "unknown arg: $arg" >&2
      exit 4
      ;;
  esac
done

# ─── helpers ────────────────────────────────────────────────────────────────
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
cd "$SCRIPT_DIR/.."   # repo root (scripts/ is a sibling of docker-compose.yml)
REPO_ROOT="$(pwd)"

if ! command -v docker >/dev/null 2>&1; then
  echo "FATAL: docker not on PATH" >&2
  exit 5
fi
if ! docker compose version >/dev/null 2>&1; then
  echo "FATAL: 'docker compose' (v2 plugin) not available — install the compose plugin" >&2
  exit 5
fi

# Timestamped log so concurrent runs (rare) don't trample each other and so
# CI can grep the timestamped output for the failure point.
LOG="$REPO_ROOT/.smoke-$(date -u +%Y%m%dT%H%M%SZ).log"
exec > >(tee -a "$LOG") 2>&1
echo "=== ci-smoke starting at $(date -u +%Y-%m-%dT%H:%M:%SZ) ==="
echo "=== repo:  $REPO_ROOT"
echo "=== log:   $LOG"
echo

cleanup() {
  local exit_code=$?
  echo
  echo "=== ci-smoke exiting with code $exit_code at $(date -u +%Y-%m-%dT%H:%M:%SZ) ==="
  if [ $KEEP_UP -eq 0 ] && [ $exit_code -ne 0 ]; then
    echo
    echo "stack left running for post-mortem. Tear down with:"
    echo "  docker compose down -v"
  fi
  if [ $KEEP_UP -eq 1 ] && [ $exit_code -eq 0 ]; then
    echo
    echo "stack left running (--keep-up). Tear down with:"
    echo "  docker compose down -v"
  fi
  exit $exit_code
}
trap cleanup EXIT INT TERM

# ─── step 1: build + start ───────────────────────────────────────────────────
if [ $SKIP_BUILD -eq 0 ]; then
  echo ">>> [1/5] docker compose build"
  docker compose build
fi

echo
echo ">>> [1/5] docker compose up -d"
docker compose up -d

# ─── step 2: wait for services healthy ───────────────────────────────────────
echo
echo ">>> [2/5] waiting for neo4j, postgres, minio to be healthy (60s deadline each)"
SERVICES=(lore-neo4j lore-postgres lore-minio)
DEADLINE_SECS=60
for svc in "${SERVICES[@]}"; do
  elapsed=0
  while [ $elapsed -lt $DEADLINE_SECS ]; do
    status=$(docker inspect -f '{{.State.Health.Status}}' "$svc" 2>/dev/null || echo "missing")
    if [ "$status" = "healthy" ]; then
      echo "  ✔ $svc healthy (after ${elapsed}s)"
      break
    fi
    if [ "$status" = "unhealthy" ]; then
      echo "  ✖ $svc reported UNHEALTHY:" >&2
      docker logs --tail 50 "$svc" >&2
      exit 1
    fi
    sleep 2
    elapsed=$((elapsed + 2))
  done
  if [ $elapsed -ge $DEADLINE_SECS ]; then
    echo "  ✖ $svc did not become healthy within ${DEADLINE_SECS}s" >&2
    echo "    last status: $status" >&2
    docker logs --tail 50 "$svc" >&2
    exit 1
  fi
done

# ─── step 3: wait for gateway /healthz ───────────────────────────────────────
echo
echo ">>> [3/5] waiting for gateway /healthz (60s deadline)"
elapsed=0
HEALTHZ_URL="${GATEWAY:-http://localhost:8765/healthz}"
while [ $elapsed -lt $DEADLINE_SECS ]; do
  if response=$(curl -fsS "$HEALTHZ_URL" 2>/dev/null) && \
     echo "$response" | python3 -c "import json,sys; d=json.loads(sys.stdin.read()); assert d.get('status')=='ok'; assert isinstance(d.get('plugins'), list) and len(d['plugins'])>0" 2>/dev/null; then
    tool_count=$(echo "$response" | python3 -c "import json,sys; print(len(json.loads(sys.stdin.read())['plugins']))")
    echo "  ✔ gateway healthy, $tool_count tools registered"
    break
  fi
  sleep 2
  elapsed=$((elapsed + 2))
done
if [ $elapsed -ge $DEADLINE_SECS ]; then
  echo "  ✖ gateway /healthz did not return 200+valid JSON within ${DEADLINE_SECS}s" >&2
  docker logs --tail 50 lore-gateway >&2 || true
  exit 1
fi

# ─── step 4: seed (idempotent — skip if data already present) ────────────────
echo
echo ">>> [4/5] seed: check if data is already loaded"
seeded_already=0
# Probe Neo4j for any Person node. If the count > 0, treat as already seeded.
# (Cheap, ~50ms.) seed.py is idempotent so re-running is safe, but skipping
# the seed keeps the smoke fast when the caller just wants to re-verify
# test.sh against a known-good DB.
person_count=$(docker exec lore-neo4j cypher-shell -u neo4j -p lore-dev-password \
  "MATCH (p:Person) RETURN count(p) AS n" 2>/dev/null \
  | awk '/^[0-9]+$/{print; exit}' || echo "0")
if [ "${person_count:-0}" -gt 0 ] 2>/dev/null; then
  echo "  ✔ already seeded (Person count = $person_count), skipping seed.py"
  seeded_already=1
fi

if [ $seeded_already -eq 0 ]; then
  echo "  → running python3 seed.py (host)"
  if ! python3 seed.py 2>/tmp/seed.err; then
    echo "  ⚠ host seed.py failed: $(head -1 /tmp/seed.err)" >&2
    echo "  → falling back to docker run via the gateway network"
    # Run seed.py inside a sidecar container on the lore-engine-poc_default
    # network. We use the gateway image because it has all the python deps,
    # then bind-mount the repo so seed.py can find the mock-data dir.
    if ! docker run --rm --network lore-engine-poc_default \
        -v "$REPO_ROOT":/work -w /work \
        -e NEO4J_URL='bolt://neo4j:7687' \
        -e NEO4J_USER=neo4j -e NEO4J_PASSWORD=lore-dev-password \
        -e POSTGRES_URL='postgresql://lore:***@postgres:5432/lore' \
        -e MINIO_URL='http://minio:9000' \
        -e MINIO_ACCESS_KEY=lorelore -e MINIO_SECRET_KEY=lore-dev-password \
        -e MINIO_BUCKET=lore-images \
        --entrypoint python3 \
        lore-engine-poc-gateway \
        seed.py 2>/tmp/seed-docker.err; then
      echo "  ✖ seed failed in both host and docker modes" >&2
      echo "    host stderr: $(cat /tmp/seed.err)" >&2
      echo "    docker stderr: $(cat /tmp/seed-docker.err)" >&2
      exit 2
    fi
  fi
  echo "  ✔ seed complete"
fi

# ─── step 5: e2e test ────────────────────────────────────────────────────────
echo
echo ">>> [5/5] bash test.sh"
if ! bash test.sh; then
  echo "  ✖ test.sh failed" >&2
  exit 3
fi
echo "  ✔ test.sh passed"

# ─── optional teardown ──────────────────────────────────────────────────────
if [ $KEEP_UP -eq 0 ]; then
  echo
  echo ">>> tearing down stack (use --keep-up to leave it running)"
  docker compose down -v
else
  echo
  echo ">>> --keep-up set, stack left running"
fi

echo
echo "=== SMOKE PASSED at $(date -u +%Y-%m-%dT%H:%M:%SZ) ==="
exit 0