#!/usr/bin/env bash
# Despliega metrics_agent en un móvil Android/Termux como servicio supervisado
# (termux-services / runit) con arranque en boot (Termux:Boot) y wake-lock.
#
# Uso: ./deploy_agent_termux.sh <node> <ssh_alias>
#
# Requisitos:
#   - Binario en apps/metrics_agent/dist/metrics_agent_arm64
#     (CGO_ENABLED=0 GOOS=linux GOARCH=arm64 go build -ldflags="-s -w" -o dist/metrics_agent_arm64 .)
#   - `pass fleet/ingest-pass`.
#   - En el móvil: Termux + Termux:Boot + Termux:API instalados, paquetes
#     `termux-services` y `termux-api`, y un alias SSH al sshd de Termux.
set -euo pipefail

NODE="${1:?uso: deploy_agent_termux.sh <node> <ssh_alias> [role]}"
HOST="${2:?uso: deploy_agent_termux.sh <node> <ssh_alias> [role]}"
ROLE="${3:-movil}"

HUB="https://metrics-dxaqj3ina6eqd5pjt85wkrrj.organic-machine.com/api/v1/import/prometheus"
LOKI="https://logs-wmaxecsjcfnocz81d5luca92.organic-machine.com/loki/api/v1/push"
PW="$(pass show fleet/ingest-pass | head -1)"
BIN="$(cd "$(dirname "$0")/.." && pwd)/apps/metrics_agent/dist/metrics_agent_arm64"

[ -f "$BIN" ] || { echo "ERROR: falta el binario $BIN (compílalo primero)"; exit 1; }

echo ">> copiando binario a $HOST"
ssh -o BatchMode=yes "$HOST" 'mkdir -p ~/fleet-agent'
scp -q -o BatchMode=yes "$BIN" "$HOST:fleet-agent/metrics_agent"

echo ">> instalando servicio Termux en $NODE ($HOST)"
ssh -o BatchMode=yes "$HOST" "NODE='$NODE' PW='$PW' HUB='$HUB' LOKI='$LOKI' ROLE='$ROLE' bash -s" <<'OUTER'
set -e
PREFIX=/data/data/com.termux/files/usr
HM=/data/data/com.termux/files/home
chmod +x "$HM/fleet-agent/metrics_agent"

cat > "$HM/fleet-agent/agent.json" <<JSON
{
  "node": "${NODE}",
  "hub_url": "${HUB}",
  "loki_url": "${LOKI}",
  "user": "fleet",
  "pass": "${PW}",
  "interval_sec": 15,
  "labels": { "role": "${ROLE}" },
  "battery_file": "$HM/fleet-agent/battery.json",
  "log_file": "$HM/fleet-agent/logcat.log"
}
JSON
chmod 600 "$HM/fleet-agent/agent.json"

# Servicio runit del agente Go (solo lee ficheros: métricas + battery.json + tail logcat.log).
mkdir -p "$PREFIX/var/service/fleet-agent"
cat > "$PREFIX/var/service/fleet-agent/run" <<RUN
#!$PREFIX/bin/sh
exec 2>&1
termux-wake-lock
exec $HM/fleet-agent/metrics_agent -config $HM/fleet-agent/agent.json
RUN
chmod +x "$PREFIX/var/service/fleet-agent/run"

# Helper shell de batería: el exec de sh SÍ funciona en Termux (el bloqueado es
# el pidfd_open de Go). Escribe el JSON de termux-battery-status cada 20s.
mkdir -p "$PREFIX/var/service/fleet-battery"
cat > "$PREFIX/var/service/fleet-battery/run" <<RUN
#!$PREFIX/bin/sh
exec 2>&1
termux-wake-lock
while true; do
  termux-battery-status > $HM/fleet-agent/battery.json.tmp 2>/dev/null && mv $HM/fleet-agent/battery.json.tmp $HM/fleet-agent/battery.json
  sleep 20
done
RUN
chmod +x "$PREFIX/var/service/fleet-battery/run"

# Helper shell de logcat: vuelca el log de Android a fichero rotado (cap ~4MB).
# El agente Go lo tail-ea sin exec.
mkdir -p "$PREFIX/var/service/fleet-logcat"
cat > "$PREFIX/var/service/fleet-logcat/run" <<RUN
#!$PREFIX/bin/sh
exec 2>&1
termux-wake-lock
exec logcat -v epoch -f $HM/fleet-agent/logcat.log -r 2048 -n 2
RUN
chmod +x "$PREFIX/var/service/fleet-logcat/run"

# Arranque en boot (Termux:Boot): wake-lock + runsvdir vía profile.
mkdir -p "$HM/.termux/boot"
cat > "$HM/.termux/boot/start-fleet.sh" <<BOOT
#!$PREFIX/bin/sh
termux-wake-lock
. $PREFIX/etc/profile
BOOT
chmod +x "$HM/.termux/boot/start-fleet.sh"

# Arrancar ya sin esperar a un reinicio: asegurar runsvdir vivo.
if ! pgrep -f "runsvdir.*var/service" >/dev/null 2>&1; then
  nohup runsvdir "$PREFIX/var/service" >/dev/null 2>&1 &
  sleep 6
fi
export SVDIR="$PREFIX/var/service"
for s in fleet-logcat fleet-battery fleet-agent; do sv up "$s" 2>/dev/null || true; done
sleep 5
for s in fleet-logcat fleet-battery fleet-agent; do echo -n "$s: "; sv status "$s" 2>/dev/null || echo "N/A"; done
OUTER

echo ">> $NODE desplegado"