#!/usr/bin/env bash
# failover.sh — graceful HA cluster failover
#
# Detects which node is active and moves all resources to the other node by
# putting the active node into Pacemaker standby.  Waits for the XFS mount to
# appear on the target before returning.
#
# Usage:
#   scripts/ha/failover.sh [--to node1|node2] [--force] [--timeout <s>] [--dry-run]
#
#   --to node1|node2  target node (default: the node that is NOT currently active)
#   --force           skip the interactive confirmation prompt
#   --timeout <s>     seconds to wait for resources to move (default: 120)
#   --dry-run         show what would be done without changing anything
set -euo pipefail

# ── Configuration ─────────────────────────────────────────────────────────
NODE1="${NODE1:-ha-server-1}"
NODE2="${NODE2:-ha-server-2}"
NODE1_IP="${NODE1_IP:-192.168.2.228}"
NODE2_IP="${NODE2_IP:-192.168.2.227}"
VIP="${VIP:-192.168.2.229}"
XFS_MOUNT="${XFS_MOUNT:-/srv/ha-data}"
HA_USER="${HA_USER:-nixos}"
# ──────────────────────────────────────────────────────────────────────────

SSH_OPTS="-i ~/.ssh/id_ed25519 -o StrictHostKeyChecking=no -o ConnectTimeout=5"

n1() { ssh $SSH_OPTS "${HA_USER}@${NODE1_IP}" sudo "$@" 2>/dev/null; }
n2() { ssh $SSH_OPTS "${HA_USER}@${NODE2_IP}" sudo "$@" 2>/dev/null; }

# ── Argument parsing ───────────────────────────────────────────────────────
TARGET_NODE=""
FORCE=false
DRY_RUN=false
TIMEOUT=120

while [[ $# -gt 0 ]]; do
  case "$1" in
    --to)
      shift
      case "${1:-}" in
        node1|ha-server-1) TARGET_NODE="$NODE1" ;;
        node2|ha-server-2) TARGET_NODE="$NODE2" ;;
        *) echo "ERROR: --to must be node1, node2, ha-server-1, or ha-server-2"; exit 1 ;;
      esac
      ;;
    --force)   FORCE=true ;;
    --dry-run) DRY_RUN=true ;;
    --timeout) shift; TIMEOUT="${1:?--timeout requires a value}" ;;
    *) echo "Unknown argument: $1"; echo "Usage: $0 [--to node1|node2] [--force] [--timeout <s>] [--dry-run]"; exit 1 ;;
  esac
  shift
done

DRY_PREFIX=""
$DRY_RUN && DRY_PREFIX="[dry-run] "

echo "════════════════════════════════════════════════════"
echo " HA Cluster Failover — $(date '+%Y-%m-%d %H:%M:%S')"
$DRY_RUN && echo " MODE: dry-run — no changes will be made"
echo "════════════════════════════════════════════════════"

# ── Detect active node ─────────────────────────────────────────────────────
echo ""
echo "Detecting active node..."

CRM_OUT=""
if ssh $SSH_OPTS "${HA_USER}@${NODE1_IP}" true 2>/dev/null; then
  CRM_OUT=$(n1 "crm_mon -1 2>/dev/null" 2>/dev/null || true)
fi
if [[ -z "$CRM_OUT" ]]; then
  if ssh $SSH_OPTS "${HA_USER}@${NODE2_IP}" true 2>/dev/null; then
    CRM_OUT=$(n2 "crm_mon -1 2>/dev/null" 2>/dev/null || true)
  fi
fi

ACTIVE_NODE=$(echo "$CRM_OUT" | grep -E '^\s*(Promoted|Masters):' | \
  grep -oE '\b(ha-server-[0-9]+)\b' | head -1 || true)

if [[ -z "$ACTIVE_NODE" ]]; then
  echo ""
  echo "ERROR: could not determine active node from crm_mon."
  echo "  Is Pacemaker still settling?  Try running scripts/ha/health.sh first."
  echo "  If Pacemaker is down on both nodes, manual recovery is required."
  exit 1
fi

if [[ "$ACTIVE_NODE" == "$NODE1" ]]; then
  ACTIVE_IP="$NODE1_IP"
  STANDBY_NODE="$NODE2"
  STANDBY_IP="$NODE2_IP"
  na() { n1 "$@"; }
  ns() { n2 "$@"; }
else
  ACTIVE_IP="$NODE2_IP"
  STANDBY_NODE="$NODE1"
  STANDBY_IP="$NODE1_IP"
  na() { n2 "$@"; }
  ns() { n1 "$@"; }
fi

echo "  Active:  $ACTIVE_NODE ($ACTIVE_IP)"
echo "  Standby: $STANDBY_NODE ($STANDBY_IP)"

# ── Validate target ────────────────────────────────────────────────────────
if [[ -n "$TARGET_NODE" ]]; then
  if [[ "$TARGET_NODE" == "$ACTIVE_NODE" ]]; then
    echo ""
    echo "ERROR: $TARGET_NODE is already the active node — nothing to do."
    exit 1
  fi
  echo "  Target:  $TARGET_NODE (as requested)"
else
  echo "  Target:  $STANDBY_NODE (auto — the other node)"
fi

# ── Pre-checks ─────────────────────────────────────────────────────────────
echo ""
echo "Pre-checks..."

DRBD_DSTATE=$(na "drbdadm dstate ha-data 2>/dev/null" 2>/dev/null || echo "unknown")
if ! echo "$DRBD_DSTATE" | grep -q "UpToDate/UpToDate"; then
  echo ""
  echo "  WARNING: DRBD dstate is '$DRBD_DSTATE' (not UpToDate/UpToDate)."
  echo "  Failing over with a partially-synced disk risks split-brain."
  if ! $FORCE; then
    echo "  Use --force to proceed anyway (not recommended)."
    exit 1
  fi
  echo "  --force specified — proceeding despite non-ideal DRBD state."
else
  echo "  DRBD dstate: $DRBD_DSTATE — OK"
fi

QUORUM_OK=$(na "corosync-quorumtool -s 2>/dev/null | grep -c 'Quorate:.*Yes'" 2>/dev/null || echo "0")
if [[ "$QUORUM_OK" -lt 1 ]]; then
  echo "  ERROR: cluster does not have quorum — failover would be unsafe."
  exit 1
fi
echo "  Quorum: OK"

# ── Confirm ────────────────────────────────────────────────────────────────
if ! $FORCE && ! $DRY_RUN; then
  echo ""
  echo "  This will move all resources from $ACTIVE_NODE → $STANDBY_NODE."
  echo "  VIP and services will be unreachable for ~10–30 seconds."
  printf "  Proceed? [y/N] "
  read -r ANSWER
  [[ "${ANSWER,,}" == "y" || "${ANSWER,,}" == "yes" ]] || { echo "Aborted."; exit 0; }
fi

# ── Capture active node's crm_node name ───────────────────────────────────
# crm_node -n returns the node name as registered in Pacemaker (may differ
# from hostname if Pacemaker was configured with explicit node names).
ACTIVE_CRMD_NAME=$(na "crm_node -n 2>/dev/null" 2>/dev/null || echo "$ACTIVE_NODE")

# ── Perform failover ───────────────────────────────────────────────────────
echo ""
echo "${DRY_PREFIX}Putting $ACTIVE_NODE into standby (resources will migrate to $STANDBY_NODE)..."
if ! $DRY_RUN; then
  na "crm_standby -N '${ACTIVE_CRMD_NAME}' -v on" 2>/dev/null || true
fi

# ── Wait for resources to move ─────────────────────────────────────────────
echo "${DRY_PREFIX}Waiting up to ${TIMEOUT}s for XFS to mount on $STANDBY_NODE..."
MOVED=false
SPIN_CHARS=('|' '/' '-' '\')
SPIN_I=0

if $DRY_RUN; then
  echo "  [dry-run] would wait for mountpoint $XFS_MOUNT on $STANDBY_NODE"
  MOVED=true
else
  for i in $(seq 1 "$TIMEOUT"); do
    if ns "mountpoint -q '${XFS_MOUNT}' 2>/dev/null" 2>/dev/null; then
      printf "\r%-80s\r" ""
      echo "  Resources moved in ${i}s"
      MOVED=true
      break
    fi
    SPIN_I=$(( SPIN_I + 1 ))
    SC="${SPIN_CHARS[$((SPIN_I % 4))]}"
    printf "\r  [%s] waiting... (%ds)     " "$SC" "$i"
    sleep 1
  done
fi

if ! $MOVED; then
  echo ""
  echo "ERROR: XFS did not mount on $STANDBY_NODE within ${TIMEOUT}s."
  echo ""
  echo "  Current resource state:"
  na "crm_mon -1 2>/dev/null" 2>/dev/null | grep -E 'Started|Stopped|Promoted|Unpromoted|FAILED' | sed 's/^/  /' || true
  echo ""
  echo "  Clearing standby to restore $ACTIVE_NODE (undo the failover attempt)..."
  na "crm_standby -N '${ACTIVE_CRMD_NAME}' -v off" 2>/dev/null || true
  na "crm_resource --cleanup" 2>/dev/null || true
  exit 1
fi

# ── Clear failure history ──────────────────────────────────────────────────
echo "${DRY_PREFIX}Clearing Pacemaker failure history..."
if ! $DRY_RUN; then
  ns "crm_resource --cleanup 2>/dev/null" 2>/dev/null || true
fi

# ── Re-enable original active node as standby ─────────────────────────────
echo "${DRY_PREFIX}Re-enabling $ACTIVE_NODE (now standby — will not claim resources)..."
if ! $DRY_RUN; then
  na "crm_standby -N '${ACTIVE_CRMD_NAME}' -v off" 2>/dev/null || true
fi

# ── Wait briefly for DRBD resync to begin ─────────────────────────────────
if ! $DRY_RUN; then
  sleep 5
fi

# ── Final state ────────────────────────────────────────────────────────────
echo ""
echo "Failover complete.  Final state:"
echo ""

CRM_OUT_AFTER=""
if ! $DRY_RUN; then
  CRM_OUT_AFTER=$(ns "crm_mon -1 2>/dev/null" 2>/dev/null || na "crm_mon -1 2>/dev/null" 2>/dev/null || true)
else
  CRM_OUT_AFTER="$CRM_OUT"
fi

NEW_ACTIVE=$(echo "$CRM_OUT_AFTER" | grep -E '^\s*(Promoted|Masters):' | \
  grep -oE '\b(ha-server-[0-9]+)\b' | head -1 || true)

if [[ -n "$NEW_ACTIVE" ]]; then
  if [[ "$NEW_ACTIVE" == "$ACTIVE_NODE" ]]; then
    echo "  WARNING: $ACTIVE_NODE is still showing as active in crm_mon."
    echo "  Pacemaker may still be settling — check again in a few seconds."
  else
    echo "  Active:  $NEW_ACTIVE"
    echo "  Standby: $ACTIVE_NODE"
  fi
fi

echo ""
RESOURCES_AFTER=$(echo "$CRM_OUT_AFTER" | awk '/Full List of Resources/,0' | tail -n +2 || true)
[[ -z "$RESOURCES_AFTER" ]] && RESOURCES_AFTER=$(echo "$CRM_OUT_AFTER" | \
  grep -E 'Started|Stopped|Promoted|Unpromoted|FAILED|Master|Slave' || true)
[[ -n "$RESOURCES_AFTER" ]] && echo "$RESOURCES_AFTER" | sed 's/^/  /'

echo ""
echo "  (DRBD resync of $ACTIVE_NODE may take a moment; monitor with:"
echo "   ssh nixos@${ACTIVE_IP} 'sudo watch -n3 cat /proc/drbd')"
echo ""
echo "════════════════════════════════════════════════════"