#!/bin/bash
# Prometheus + Grafana for the labsim connectivity matrix.
#
# Grafana runs with anonymous auth as Admin — NO LOGIN. That is deliberate for
# a throwaway lab on localhost; do not copy this into anything reachable.
#
#   ./monitoring-up.sh          start exporter + prometheus + grafana
#   ./monitoring-up.sh --down   stop and remove them
#
# Grafana:    http://localhost:3000   (dashboard "labsim — VLAN connectivity matrix")
# Prometheus: http://localhost:9090
# Exporter:   http://localhost:9101/metrics
set -euo pipefail

SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
source "$SCRIPT_DIR/lib.sh"

GRAFANA_PORT="${GRAFANA_PORT:-3000}"
PROM_PORT="${PROM_PORT:-9090}"
EXPORTER_PORT="${EXPORTER_PORT:-9101}"
NET="labsim-mon"

if [ "${1:-}" = "--down" ]; then
  pkill -f "labsim-exporter.py" 2>/dev/null || true
  podman rm -f labsim-grafana labsim-prometheus >/dev/null 2>&1 || true
  podman network rm -f "$NET" >/dev/null 2>&1 || true
  log "monitoring stopped"
  exit 0
fi

command -v podman >/dev/null 2>&1 || die "podman not installed"

# --- exporter (on the host: it needs SSH access to the VMs) ----------------
if pgrep -f "labsim-exporter.py" >/dev/null 2>&1; then
  log "exporter already running on :$EXPORTER_PORT"
else
  log "starting exporter on :$EXPORTER_PORT"
  nohup "$SCRIPT_DIR/labsim-exporter.py" --port "$EXPORTER_PORT" --interval 15 \
    > /tmp/labsim-exporter.log 2>&1 &
  sleep 3
fi
curl -sS --max-time 5 "http://127.0.0.1:${EXPORTER_PORT}/metrics" >/dev/null \
  || die "exporter not answering on :$EXPORTER_PORT (see /tmp/labsim-exporter.log)"

podman network exists "$NET" 2>/dev/null || podman network create "$NET" >/dev/null

# --- prometheus -----------------------------------------------------------
podman rm -f labsim-prometheus >/dev/null 2>&1 || true
log "starting prometheus on :$PROM_PORT"
podman run -d --name labsim-prometheus --network "$NET" \
  -p "${PROM_PORT}:9090" \
  -v "$SCRIPT_DIR/monitoring/prometheus.yml:/etc/prometheus/prometheus.yml:ro,Z" \
  --add-host "host.containers.internal:host-gateway" \
  docker.io/prom/prometheus:latest >/dev/null

# --- grafana (anonymous, no login) ----------------------------------------
podman rm -f labsim-grafana >/dev/null 2>&1 || true
log "starting grafana on :$GRAFANA_PORT (anonymous auth — no password)"
podman run -d --name labsim-grafana --network "$NET" \
  -p "${GRAFANA_PORT}:3000" \
  -e GF_AUTH_ANONYMOUS_ENABLED=true \
  -e GF_AUTH_ANONYMOUS_ORG_ROLE=Admin \
  -e GF_AUTH_DISABLE_LOGIN_FORM=true \
  -e GF_AUTH_BASIC_ENABLED=false \
  -e GF_SECURITY_ALLOW_EMBEDDING=true \
  -e GF_USERS_DEFAULT_THEME=dark \
  -v "$SCRIPT_DIR/monitoring/grafana/provisioning:/etc/grafana/provisioning:ro,Z" \
  docker.io/grafana/grafana:latest >/dev/null

log "waiting for grafana..."
for _ in $(seq 1 40); do
  if curl -sS --max-time 3 "http://127.0.0.1:${GRAFANA_PORT}/api/health" >/dev/null 2>&1; then
    break
  fi
  sleep 3
done

echo
log "Grafana:    http://localhost:${GRAFANA_PORT}/d/labsim-matrix  (no login)"
log "Prometheus: http://localhost:${PROM_PORT}"
log "Exporter:   http://localhost:${EXPORTER_PORT}/metrics"