#!/bin/bash
# cluster-gateway.sh — M3U96a comme porte d'entrée (staging) du cluster MacLustr.
#
# Principe : on pousse un projet UNE fois laptop -> M3U96a (~/cluster-projects/<nom>/),
# puis M3U96a le diffuse aux autres noeuds en parallele via le reseau local (Tailscale/LAN).
# Le meme chemin ~/cluster-projects/<nom>/ existe ainsi sur tous les noeuds cibles.
#
# Usage :
#   cluster-gateway.sh stage  <dir-local> [nom]             # laptop -> M3U96a
#   cluster-gateway.sh fanout <nom> [noeud ...]             # M3U96a -> noeuds (defaut: tous sauf M3U96a & M2U64)
#   cluster-gateway.sh push   <dir-local> [nom]             # stage + fanout (defaut)
#   cluster-gateway.sh list                                 # projets stages sur M3U96a
#   cluster-gateway.sh where  <nom>                         # sur quels noeuds le projet est present
#   cluster-gateway.sh clean  <nom> [--everywhere]          # efface de M3U96a (et de tous les noeuds)
#
# Options : --include-m2u64 (fanout/push : inclure le noeud prod M2U64)
# Doc complete : ~/Desktop/cluster-skill/GATEWAY.md
set -euo pipefail

GATEWAY="M3U96a"
DOMAIN="maclustr.io"
STAGE_DIR="cluster-projects"
LAN_PREFIX_RE='192\.168\.2\.'   # sous-reseau LAN du cluster
ALL_NODES="M3U96b M2U64 M4M64a M4M64b M4BP48 M4BP36 M4M36 M2M32 M2M32b M2M32c m4mc M1M32 m4ma m4mb m2m16 M3BA24 M3BA16 m2m8a m2m8b"

usage() { sed -n '2,18p' "$0" | sed 's/^# \{0,1\}//'; exit 1; }

CMD="${1:-}"; shift || true

INCLUDE_M2U64=0
ARGS=()
for a in "$@"; do
  case "$a" in
    --include-m2u64) INCLUDE_M2U64=1 ;;
    *) ARGS+=("$a") ;;
  esac
done
set -- "${ARGS[@]:-}"

default_targets() {
  local out=""
  for n in $ALL_NODES; do
    [ "$n" = "M2U64" ] && [ "$INCLUDE_M2U64" -eq 0 ] && continue
    out="$out $n"
  done
  echo "$out"
}

do_stage() {
  local src="${1:?usage: stage <dir-local> [nom]}"
  src="${src%/}"
  [ -d "$src" ] || { echo "ERREUR: $src n'est pas un dossier" >&2; exit 2; }
  local name="${2:-$(basename "$src")}"
  echo ">> stage: $src -> $GATEWAY:~/$STAGE_DIR/$name/"
  rsync -az --delete --rsync-path="mkdir -p $STAGE_DIR && rsync" \
    "$src/" "$GATEWAY:$STAGE_DIR/$name/"
  echo "OK  stage $name ($(du -sh "$src" | cut -f1))"
}

do_fanout() {
  local name="${1:?usage: fanout <nom> [noeud ...]}"; shift || true
  local nodes="${*:-$(default_targets)}"
  # Le SSH inter-noeuds via Tailscale est filtre (ACL) : le fanout passe par le LAN
  # 192.168.2.x. Les IP LAN sont en DHCP -> on les resout en live depuis le laptop.
  echo ">> resolution des IP LAN des cibles ($(echo $nodes | wc -w | tr -d ' ') noeuds)..."
  local tmp; tmp=$(mktemp -d)
  local n
  for n in $nodes; do
    ( ip=$(ssh -n -o ConnectTimeout=8 "$n" \
        "ifconfig | awk '/inet $LAN_PREFIX_RE/{print \$2; exit}'" 2>/dev/null || true)
      echo "${ip:-NONE}" >"$tmp/$n" ) &
  done
  wait
  local pairs=""
  for n in $nodes; do pairs="$pairs $n=$(cat "$tmp/$n" 2>/dev/null || echo NONE)"; done
  rm -rf "$tmp"
  echo ">> fanout depuis $GATEWAY via LAN: $name ->$(echo "$pairs" | tr ' ' '\n' | grep -c '=' | xargs echo) noeuds"
  # shellcheck disable=SC2087
  ssh -A "$GATEWAY" "NAME='$name' PAIRS='$pairs' bash -s" <<'REMOTE'
set -u
SRC="$HOME/cluster-projects/$NAME/"
[ -d "$SRC" ] || { echo "ERREUR: $SRC absent sur la passerelle — faire stage d abord" >&2; exit 2; }
TMP=$(mktemp -d)
for pair in $PAIRS; do
  n="${pair%%=*}"; ip="${pair##*=}"
  if [ "$ip" = "NONE" ]; then echo "FAIL (hors LAN cluster ou injoignable)" >"$TMP/$n"; continue; fi
  ( rsync -az --delete \
      -e "ssh -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null -o ConnectTimeout=8" \
      --rsync-path="mkdir -p cluster-projects && rsync" \
      "$SRC" "$ip:cluster-projects/$NAME/" \
      >/dev/null 2>"$TMP/$n.err" && echo "OK ($ip)" >"$TMP/$n" || echo "FAIL ($ip)" >"$TMP/$n" ) &
done
wait
FAILED=0
for pair in $PAIRS; do
  n="${pair%%=*}"
  st=$(cat "$TMP/$n" 2>/dev/null || echo FAIL)
  printf "  %-8s %s\n" "$n" "$st"
  case "$st" in FAIL*) FAILED=1; [ -f "$TMP/$n.err" ] && sed 's/^/           /' "$TMP/$n.err" | grep -v "known hosts" | head -2 ;; esac
done
rm -rf "$TMP"
exit $FAILED
REMOTE
  echo "OK  fanout $name termine"
}

do_list() {
  echo ">> projets stages sur $GATEWAY:~/$STAGE_DIR/"
  ssh "$GATEWAY" "du -sh $STAGE_DIR/*/ 2>/dev/null || echo '  (aucun projet stage)'"
}

do_where() {
  local name="${1:?usage: where <nom>}"
  echo ">> presence de '$name' sur le cluster :"
  for n in $GATEWAY $ALL_NODES; do
    ( sz=$(ssh -o ConnectTimeout=6 "$n.$DOMAIN" "du -sh $STAGE_DIR/$name 2>/dev/null | cut -f1" 2>/dev/null)
      printf "  %-8s %s\n" "$n" "${sz:-—}" ) &
  done
  wait
}

do_clean() {
  local name="${1:?usage: clean <nom> [--everywhere]}"
  local everywhere=0
  [ "${2:-}" = "--everywhere" ] && everywhere=1
  if [ "$everywhere" -eq 1 ]; then
    echo ">> clean '$name' sur TOUS les noeuds"
    for n in $GATEWAY $ALL_NODES; do
      ( ssh -o ConnectTimeout=6 "$n.$DOMAIN" "rm -rf $STAGE_DIR/$name" 2>/dev/null \
          && printf "  %-8s OK\n" "$n" || printf "  %-8s (injoignable)\n" "$n" ) &
    done
    wait
  else
    echo ">> clean '$name' sur $GATEWAY uniquement"
    ssh "$GATEWAY" "rm -rf $STAGE_DIR/$name" && echo "  $GATEWAY OK"
  fi
}

case "$CMD" in
  stage)  do_stage "$@" ;;
  fanout) do_fanout "$@" ;;
  push)   do_stage "$1" "${2:-}"; do_fanout "${2:-$(basename "${1%/}")}" ;;
  list)   do_list ;;
  where)  do_where "$@" ;;
  clean)  do_clean "$@" ;;
  *)      usage ;;
esac