#!/usr/bin/env bash
set -euo pipefail
DOCKER_SMOKE_RUN_ID="${DOCKER_SMOKE_RUN_ID:-$$}"
if [[ -z "$DOCKER_SMOKE_RUN_ID" || ${#DOCKER_SMOKE_RUN_ID} -gt 80 || ! $DOCKER_SMOKE_RUN_ID =~ ^[A-Za-z0-9_.-]+$ ]]; then
  echo "DOCKER_SMOKE_RUN_ID must contain 1-80 Docker-safe characters" >&2
  exit 2
fi

# shellcheck source=scripts/lib/docker-rls-role-fixture.sh
source "$(dirname "${BASH_SOURCE[0]}")/lib/docker-rls-role-fixture.sh"
# shellcheck source=scripts/lib/docker-runtime-smoke.sh
source "$(dirname "${BASH_SOURCE[0]}")/lib/docker-runtime-smoke.sh"

IMAGE="${1:-hivegrid-mcp:smoke}"
MIGRATOR_IMAGE="${DOCKER_SMOKE_MIGRATOR_IMAGE:-hivegrid-web:smoke}"
PORT="${DOCKER_MCP_SMOKE_PORT:-}"
CONTAINER_NAME="${DOCKER_MCP_SMOKE_CONTAINER:-hivegrid-mcp-runtime-smoke-${DOCKER_SMOKE_RUN_ID}}"
REDIS_CONTAINER_NAME="${DOCKER_MCP_SMOKE_REDIS_CONTAINER:-hivegrid-mcp-runtime-smoke-redis-${DOCKER_SMOKE_RUN_ID}}"
DB_CONTAINER_NAME="${DOCKER_MCP_SMOKE_DB_CONTAINER:-hivegrid-mcp-runtime-smoke-db-${DOCKER_SMOKE_RUN_ID}}"
NETWORK_NAME="${DOCKER_MCP_SMOKE_NETWORK:-hivegrid-mcp-runtime-smoke-${DOCKER_SMOKE_RUN_ID}}"
MIGRATION_CONTAINER_NAME="${DOCKER_MCP_SMOKE_MIGRATION_CONTAINER:-hivegrid-mcp-runtime-smoke-migrate-${DOCKER_SMOKE_RUN_ID}}"
ASSERT_CONTAINER_NAME="${DOCKER_MCP_SMOKE_ASSERT_CONTAINER:-hivegrid-mcp-runtime-smoke-assert-${DOCKER_SMOKE_RUN_ID}}"
TRANSPORT="${DOCKER_MCP_SMOKE_TRANSPORT:-host}"
EXPECTED_SMOKE_REVISION="${DOCKER_MCP_SMOKE_REVISION:-}"
SMOKE_REVISION=""
HEALTH_ATTEMPTS="${DOCKER_MCP_SMOKE_HEALTH_ATTEMPTS:-120}"
DEPENDENCY_ATTEMPTS="${DOCKER_MCP_SMOKE_DEPENDENCY_ATTEMPTS:-30}"
DB_PASSWORD="hivegrid"
DB_NAME="hivegrid"
DB_USER="hivegrid"
APP_DB_PASSWORD="hivegrid-app-smoke"
WORKER_DB_PASSWORD="hivegrid-worker-smoke"
MIGRATOR_DB_PASSWORD="hivegrid-migrator-smoke"
AUTHENTIK_JIT_DB_PASSWORD="hivegrid-authentik-jit-smoke"

cleanup() {
  docker_smoke_cleanup_container "mcp:cleanup-containers" \
    "$CONTAINER_NAME" "$MIGRATION_CONTAINER_NAME" "$ASSERT_CONTAINER_NAME" \
    "$REDIS_CONTAINER_NAME" "$DB_CONTAINER_NAME"
  docker_smoke_cleanup_network "mcp:cleanup-network" "$NETWORK_NAME"
}

resolve_smoke_revision() {
  local image_revision

  docker_smoke_capture image_revision "mcp:inspect-revision" "$DOCKER_SMOKE_PROBE_TIMEOUT_SECONDS" \
    docker image inspect -f '{{ index .Config.Labels "org.opencontainers.image.revision" }}' "$IMAGE"
  if [[ -n "$EXPECTED_SMOKE_REVISION" && "$EXPECTED_SMOKE_REVISION" != "$image_revision" ]]; then
    echo "Image ${IMAGE} revision ${image_revision:-<empty>} does not match expected revision ${EXPECTED_SMOKE_REVISION}" >&2
    return 1
  fi

  printf "%s" "$image_revision"
}

verify_migrator_revision() {
  local migrator_revision

  docker_smoke_capture migrator_revision "mcp:inspect-migrator-revision" "$DOCKER_SMOKE_PROBE_TIMEOUT_SECONDS" \
    docker image inspect -f '{{ index .Config.Labels "org.opencontainers.image.revision" }}' "$MIGRATOR_IMAGE"
  if [[ "$migrator_revision" != "$SMOKE_REVISION" ]]; then
    echo "Migrator image ${MIGRATOR_IMAGE} revision ${migrator_revision:-<empty>} does not match MCP image revision ${SMOKE_REVISION}" >&2
    exit 1
  fi
}

http_status() {
  local path="$1"
  local method="${2:-GET}"
  local code

  if [[ "$TRANSPORT" == "exec" ]]; then
    if ! docker_smoke_capture code "mcp:http-status" "$DOCKER_SMOKE_PROBE_TIMEOUT_SECONDS" \
      docker exec "$CONTAINER_NAME" node -e 'const http = require("node:http"); const [path, method] = process.argv.slice(1); const request = http.request({ hostname: "127.0.0.1", port: 8080, path, method, headers: { Host: "mcp.hivegrid.de" } }, (response) => { process.stdout.write(String(response.statusCode ?? "000")); response.resume(); }); request.on("error", () => process.stdout.write("000")); request.end();' "$path" "$method" 2>/dev/null; then
      code="000"
    fi
  else
    if ! docker_smoke_capture code "mcp:http-status" "$DOCKER_SMOKE_PROBE_TIMEOUT_SECONDS" \
      curl -s -o /dev/null -w "%{http_code}" -X "$method" -H "Host: mcp.hivegrid.de" "http://127.0.0.1:${PORT}${path}" 2>/dev/null; then
      code="000"
    fi
  fi

  if [[ -z "$code" ]]; then
    printf "000"
    return
  fi

  printf "%s" "$code"
}

http_body() {
  local path="$1"
  local body

  if [[ "$TRANSPORT" == "exec" ]]; then
    docker_smoke_capture body "mcp:http-body" "$DOCKER_SMOKE_PROBE_TIMEOUT_SECONDS" \
      docker exec "$CONTAINER_NAME" node -e 'const http = require("node:http"); const [path] = process.argv.slice(1); const request = http.request({ hostname: "127.0.0.1", port: 8080, path, headers: { Host: "mcp.hivegrid.de" } }, (response) => { let body = ""; response.setEncoding("utf8"); response.on("data", (chunk) => { body += chunk; }); response.on("end", () => process.stdout.write(body)); }); request.on("error", () => process.exit(1)); request.end();' "$path"
  else
    docker_smoke_capture body "mcp:http-body" "$DOCKER_SMOKE_PROBE_TIMEOUT_SECONDS" \
      curl --fail --silent --show-error -H "Host: mcp.hivegrid.de" "http://127.0.0.1:${PORT}${path}"
  fi

  printf "%s" "$body"
}

expect_status() {
  local path="$1"
  local expected="$2"
  local method="${3:-GET}"
  local actual

  actual="$(http_status "$path" "$method")"
  if [[ "$actual" != "$expected" ]]; then
    echo "Expected ${method} ${path} to return ${expected}, got ${actual}" >&2
    docker_smoke_logs "mcp:logs-status" "$CONTAINER_NAME" >&2 || true
    exit 1
  fi
}

wait_for_status() {
  local path="$1"
  local expected="$2"
  local description="$3"

  for attempt in $(seq 1 "$DEPENDENCY_ATTEMPTS"); do
    if [[ "$(http_status "$path")" == "$expected" ]]; then
      return
    fi
    sleep 1
  done

  echo "MCP ${description} did not reach HTTP ${expected} within ${DEPENDENCY_ATTEMPTS}s" >&2
  docker_smoke_logs "mcp:logs-readiness" "$CONTAINER_NAME" >&2 || true
  exit 1
}

expect_health_revision() {
  local body
  local actual
  local service

  body="$(http_body /healthz)"
  actual="$(printf "%s" "$body" | node -e 'const fs = require("node:fs"); const payload = JSON.parse(fs.readFileSync(0, "utf8")); process.stdout.write(payload.revision || "");')"
  service="$(printf "%s" "$body" | node -e 'const fs = require("node:fs"); const payload = JSON.parse(fs.readFileSync(0, "utf8")); process.stdout.write(payload.service || "");')"
  if [[ "$actual" != "$SMOKE_REVISION" || "$service" != "mcp" ]]; then
    echo "Expected /healthz service=mcp revision=${SMOKE_REVISION}, got service=${service:-<empty>} revision=${actual:-<empty>}" >&2
    docker_smoke_logs "mcp:logs-revision" "$CONTAINER_NAME" >&2 || true
    exit 1
  fi
}

expect_deploy_health() {
  local output="$1"

  printf "%s" "$output" | node -e '
const expectedRevision = process.argv[1];
const payload = JSON.parse(require("node:fs").readFileSync(0, "utf8"));
if (payload.status !== "ready" || payload.service !== "mcp" || payload.revision !== expectedRevision) {
  console.error("MCP deployment health did not prove the expected live revision");
  process.exit(1);
}
' "$SMOKE_REVISION"
}

docker_smoke_install_cleanup_traps cleanup
cleanup
SMOKE_REVISION="$(resolve_smoke_revision)"
if [[ -z "$SMOKE_REVISION" || "$SMOKE_REVISION" == "<no value>" || "$SMOKE_REVISION" == "unknown" ]]; then
  echo "Image ${IMAGE} does not expose a concrete org.opencontainers.image.revision; build with --build-arg APP_REVISION" >&2
  exit 1
fi
verify_migrator_revision

docker_smoke_run "mcp:create-network" "$DOCKER_SMOKE_COMMAND_TIMEOUT_SECONDS" \
  docker network create "$NETWORK_NAME" >/dev/null
docker_smoke_run "mcp:start-redis" "$DOCKER_SMOKE_COMMAND_TIMEOUT_SECONDS" \
  docker run -d --name "$REDIS_CONTAINER_NAME" --network "$NETWORK_NAME" --network-alias redis redis:7-alpine >/dev/null
docker_smoke_run "mcp:start-db" "$DOCKER_SMOKE_COMMAND_TIMEOUT_SECONDS" docker run -d --name "$DB_CONTAINER_NAME" --network "$NETWORK_NAME" --network-alias db \
  -e POSTGRES_USER="$DB_USER" \
  -e POSTGRES_PASSWORD="$DB_PASSWORD" \
  -e POSTGRES_DB="$DB_NAME" \
  postgres:16-alpine >/dev/null

MIGRATOR_DATABASE_URL="postgresql://hivegrid_migrator:${MIGRATOR_DB_PASSWORD}@db:5432/${DB_NAME}?schema=public"
APP_DATABASE_URL="postgresql://hivegrid_app:${APP_DB_PASSWORD}@db:5432/${DB_NAME}?schema=public"

for attempt in $(seq 1 30); do
  if docker_smoke_probe "mcp:probe-redis" docker exec "$REDIS_CONTAINER_NAME" redis-cli ping 2>/dev/null | grep -q PONG; then
    break
  fi

  if [[ "$attempt" == "30" ]]; then
    echo "Redis smoke service did not become ready" >&2
    docker_smoke_logs "mcp:logs-redis" "$REDIS_CONTAINER_NAME" >&2 || true
    exit 1
  fi

  sleep 1
done

# Probe over TCP, not the unix socket. On first boot the postgres image starts a
# temporary server that listens on the socket ONLY (initdb + init scripts), then
# shuts it down before starting the real one. A socket-based pg_isready reports
# ready during that window, so the very next psql can hit
# "FATAL: the database system is shutting down". The temp server sets
# listen_addresses='', so a TCP probe cannot succeed too early.
if ! docker_smoke_wait_for_postgres "mcp:probe-db" \
  "$DB_CONTAINER_NAME" "$DB_USER" "$DB_NAME"; then
  echo "Postgres smoke database did not become ready within the ${DOCKER_SMOKE_DATABASE_READY_TIMEOUT_SECONDS}-second budget" >&2
  docker_smoke_logs "mcp:logs-db" "$DB_CONTAINER_NAME" >&2 || true
  exit 1
fi

provision_docker_smoke_rls_roles \
  "mcp:provision-rls-roles" \
  "$DB_CONTAINER_NAME" "$DB_USER" "$DB_NAME" \
  "$MIGRATOR_DB_PASSWORD" "$APP_DB_PASSWORD" "$WORKER_DB_PASSWORD" "$AUTHENTIK_JIT_DB_PASSWORD"

# Exercise the same reviewed migration closure as deployment. MCP readiness
# queries both credential tables, so an unmigrated database must stay unready.
docker_smoke_run "mcp:migrate" "$DOCKER_SMOKE_MIGRATION_TIMEOUT_SECONDS" docker run --rm --name "$MIGRATION_CONTAINER_NAME" --network "$NETWORK_NAME" \
  -e DIRECT_URL="$MIGRATOR_DATABASE_URL" \
  "$MIGRATOR_IMAGE" migrate

# MCP instances never migrate schemas or compile source. Keep build-only
# toolchains out of the remotely reachable runtime closure.
docker_smoke_run "mcp:assert-runtime" "$DOCKER_SMOKE_COMMAND_TIMEOUT_SECONDS" docker run --rm --name "$ASSERT_CONTAINER_NAME" --entrypoint node "$IMAGE" -e '
for (const specifier of ["prisma", "typescript", "@prisma/engines"]) {
  try {
    require.resolve(specifier);
    throw new Error(`MCP runtime retained build-only package ${specifier}`);
  } catch (error) {
    if (error?.code !== "MODULE_NOT_FOUND") throw error;
  }
}
' >/dev/null

publish_args=(-p 127.0.0.1::8080)
if [[ -n "$PORT" ]]; then
  publish_args=(-p "${PORT}:8080")
fi

docker_smoke_run "mcp:start-runtime" "$DOCKER_SMOKE_COMMAND_TIMEOUT_SECONDS" docker run -d --name "$CONTAINER_NAME" --network "$NETWORK_NAME" "${publish_args[@]}" \
  -e NODE_ENV=production \
  -e PORT=8080 \
  -e MCP_PORT=8080 \
  -e DATABASE_URL="$APP_DATABASE_URL" \
  -e DATABASE_RUNTIME_ROLE=hivegrid_app \
  -e RLS_RUNTIME_MODE=required \
  -e REDIS_URL='redis://redis:6379/0' \
  -e MCP_OAUTH_TOKEN_HASH_SECRET="smoke-only-mcp-token-hash-secret-not-real" \
  -e AUDIT_CHAIN_SECRET='deploy-smoke-audit-chain-secret-minimum-32' \
  "$IMAGE" >/dev/null

if [[ -z "$PORT" ]]; then
  docker_smoke_capture PORT "mcp:resolve-port" "$DOCKER_SMOKE_PROBE_TIMEOUT_SECONDS" \
    docker port "$CONTAINER_NAME" 8080/tcp
  PORT="$(printf "%s\n" "$PORT" | awk -F: 'NR == 1 { print $NF }')"
fi

for attempt in $(seq 1 "$HEALTH_ATTEMPTS"); do
  if [[ "$(http_status /healthz)" == "200" ]]; then
    break
  fi

  if [[ "$attempt" == "$HEALTH_ATTEMPTS" ]]; then
    echo "MCP container did not become healthy on port ${PORT} after ${HEALTH_ATTEMPTS}s" >&2
    docker_smoke_logs "mcp:logs-health" "$CONTAINER_NAME" >&2 || true
    exit 1
  fi

  sleep 1
done

expect_status /healthz 200
expect_health_revision
wait_for_status /readyz 200 "initial dependency readiness"
docker_smoke_run "mcp:assert-revision-file" "$DOCKER_SMOKE_COMMAND_TIMEOUT_SECONDS" docker exec "$CONTAINER_NAME" node -e '
const fs = require("node:fs");
const expectedRevision = process.argv[1];
const path = "/myapp/.image-revision";
const stat = fs.statSync(path);
const actualRevision = fs.readFileSync(path, "utf8").trim();
if (stat.uid !== 0 || (stat.mode & 0o022) !== 0 || actualRevision !== expectedRevision) {
  console.error("MCP immutable image revision file is not root-owned, read-only, and exact");
  process.exit(1);
}
' "$SMOKE_REVISION"
deploy_health_output=""
docker_smoke_capture deploy_health_output "mcp:deploy-health" "$DOCKER_SMOKE_PROBE_TIMEOUT_SECONDS" \
  docker exec "$CONTAINER_NAME" node apps/mcp/dist/deploy-health.js "$SMOKE_REVISION"
expect_deploy_health "$deploy_health_output"
MISMATCHED_REVISION="$(printf '0%.0s' {1..40})"
if [[ "$MISMATCHED_REVISION" == "$SMOKE_REVISION" ]]; then
  MISMATCHED_REVISION="$(printf 'f%.0s' {1..40})"
fi
docker_smoke_expect_docker_exec_rejection "mcp:reject-spoofed-revision" "$DOCKER_SMOKE_PROBE_TIMEOUT_SECONDS" "$CONTAINER_NAME" \
  env APP_REVISION="$MISMATCHED_REVISION" \
  node apps/mcp/dist/deploy-health.js "$MISMATCHED_REVISION"
expect_status /mcp 404 POST
expect_status / 401 POST

# Readiness must reflect the exact Redis and Postgres dependencies used by
# authenticated MCP traffic while liveness remains process-only.
docker_smoke_run "mcp:stop-redis" "$DOCKER_SMOKE_COMMAND_TIMEOUT_SECONDS" docker stop "$REDIS_CONTAINER_NAME" >/dev/null
wait_for_status /readyz 503 "Redis outage readiness"
expect_status /healthz 200
docker_smoke_expect_docker_exec_rejection "mcp:reject-redis-outage" "$DOCKER_SMOKE_PROBE_TIMEOUT_SECONDS" "$CONTAINER_NAME" \
  node apps/mcp/dist/deploy-health.js "$SMOKE_REVISION"

docker_smoke_run "mcp:start-redis-recovery" "$DOCKER_SMOKE_COMMAND_TIMEOUT_SECONDS" docker start "$REDIS_CONTAINER_NAME" >/dev/null
for attempt in $(seq 1 30); do
  if docker_smoke_probe "mcp:probe-redis-recovery" docker exec "$REDIS_CONTAINER_NAME" redis-cli ping 2>/dev/null | grep -q PONG; then
    break
  fi
  sleep 1
done
wait_for_status /readyz 200 "Redis recovery readiness"
docker_smoke_capture deploy_health_output "mcp:deploy-health-recovery" "$DOCKER_SMOKE_PROBE_TIMEOUT_SECONDS" \
  docker exec "$CONTAINER_NAME" node apps/mcp/dist/deploy-health.js "$SMOKE_REVISION"
expect_deploy_health "$deploy_health_output"

docker_smoke_run "mcp:stop-db" "$DOCKER_SMOKE_COMMAND_TIMEOUT_SECONDS" docker stop "$DB_CONTAINER_NAME" >/dev/null
wait_for_status /readyz 503 "Postgres outage readiness"
expect_status /healthz 200
docker_smoke_expect_docker_exec_rejection "mcp:reject-db-outage" "$DOCKER_SMOKE_PROBE_TIMEOUT_SECONDS" "$CONTAINER_NAME" \
  node apps/mcp/dist/deploy-health.js "$SMOKE_REVISION"

echo "Docker MCP runtime smoke passed for ${IMAGE}"