#!/bin/bash
# SPDX-License-Identifier: GPL-3.0-or-later
# zfs-replicate - Replicate ZFS datasets to TrueNAS
#
# Usage:
#   zfs-replicate              # Replicate all configured datasets
#   zfs-replicate [dataset]    # Replicate specific dataset

set -e

# TrueNAS Configuration
# Try local network first, fall back to tailscale
TRUENAS_LOCAL="truenas.local"
TRUENAS_TAILSCALE="truenas"
TRUENAS_USER="root"
TRUENAS_POOL="vault"
BACKUP_PATH="backups"  # TODO: Configure actual path

# Datasets to replicate
DATASETS="zroot/ROOT/default zroot/home zroot/media zroot/vms"

# Colors
GREEN='\033[0;32m'
YELLOW='\033[1;33m'
RED='\033[0;31m'
NC='\033[0m'

# Diagnostics go to stderr, never stdout. determine_host below runs inside a
# command substitution, so anything these print on stdout would be captured
# into TRUENAS_HOST instead of reaching the terminal or the journal -- which is
# exactly how an unreachable TrueNAS used to abort this script with exit 1 and
# no output at all. stderr also keeps the captured hostname clean.
info()  { echo -e "${GREEN}[INFO]${NC} $1" >&2; }
warn()  { echo -e "${YELLOW}[WARN]${NC} $1" >&2; }
error() { echo -e "${RED}[ERROR]${NC} $1" >&2; exit 1; }

command -v syncoid >/dev/null 2>&1 || error "syncoid not found. Install sanoid package."

# Determine which host to use
determine_host() {
    if ping -c 1 -W 2 "$TRUENAS_LOCAL" &>/dev/null; then
        echo "$TRUENAS_LOCAL"
    elif ping -c 1 -W 2 "$TRUENAS_TAILSCALE" &>/dev/null; then
        echo "$TRUENAS_TAILSCALE"
    else
        error "Cannot reach TrueNAS at $TRUENAS_LOCAL or $TRUENAS_TAILSCALE"
    fi
}

TRUENAS_HOST=$(determine_host)
info "Using TrueNAS host: $TRUENAS_HOST"

# Single dataset mode
if [[ -n "$1" ]]; then
    dataset="$1"
    dest="$TRUENAS_USER@$TRUENAS_HOST:$TRUENAS_POOL/$BACKUP_PATH/${dataset#zroot/}"
    info "Replicating $dataset -> $dest"
    syncoid --recursive "$dataset" "$dest"
    exit 0
fi

# Full replication
info "Starting ZFS replication to $TRUENAS_HOST"
echo ""

# A failed dataset must not stop the others -- but it must not be forgotten
# either. This runs as a systemd oneshot on a nightly timer, so the exit code
# is the only signal anyone sees; exiting 0 after every dataset failed made a
# backup that had not run in months look identical to a working one.
failed=0

for dataset in $DATASETS; do
    dest="$TRUENAS_USER@$TRUENAS_HOST:$TRUENAS_POOL/$BACKUP_PATH/${dataset#zroot/}"
    info "Replicating $dataset -> $dest"

    if syncoid --recursive "$dataset" "$dest"; then
        info "  Success"
    else
        warn "  Failed (will retry next run)"
        failed=$((failed + 1))
    fi
    echo "" >&2
done

if [ "$failed" -gt 0 ]; then
    warn "Replication complete with $failed of $(echo "$DATASETS" | wc -w) datasets failed."
    exit 1
fi

info "Replication complete."