#!/usr/bin/env bash
# =============================================================================
# lilith-backup-postgres.sh
# =============================================================================
# Backs up all 16 Lilith Platform PostgreSQL databases using pg_dump --format=custom.
#
# STORAGE
#   Hot: /tank/backups/lilith/postgres/daily/YYYY-MM-DD_HH/
#   Retention: 7 days (pruned at end of each run)
#
# CONFIG FILE
#   /var/home/lilith/.config/lilith/backup-postgres.conf
#   Format — one entry per line:
#     port:database:user:password
#   Example:
#     25432:lilith_prod:postgres:s3cr3t
#
# RESTORE EXAMPLE
#   # List available backups
#   ls /tank/backups/lilith/postgres/daily/
#
#   # Restore a specific database (e.g. lilith_prod from 2026-02-18_06)
#   export PGPASSWORD='s3cr3t'
#   pg_restore \
#     --host=localhost \
#     --port=25432 \
#     --username=postgres \
#     --dbname=lilith_prod \
#     --clean \
#     --if-exists \
#     /tank/backups/lilith/postgres/daily/2026-02-18_06/lilith_prod_25432.dump
#
# EXIT CODE
#   0 — all databases backed up successfully
#   1 — one or more databases failed (all are still attempted)
# =============================================================================

set -uo pipefail

# ---------------------------------------------------------------------------
# Constants
# ---------------------------------------------------------------------------
readonly BACKUP_BASE="/tank/backups/lilith/postgres/daily"
readonly CONF_FILE="/var/home/lilith/.config/lilith/backup-postgres.conf"
readonly RETENTION_DAYS=7
TIMESTAMP="$(date --utc '+%Y-%m-%d_%H')"
readonly TIMESTAMP
BACKUP_DIR="${BACKUP_BASE}/${TIMESTAMP}"
readonly BACKUP_DIR
readonly LOG_PREFIX="[lilith-backup-postgres]"

# ---------------------------------------------------------------------------
# Helpers
# ---------------------------------------------------------------------------
log()  { echo "${LOG_PREFIX} $(date --utc '+%Y-%m-%dT%H:%M:%SZ') INFO  $*"; }
warn() { echo "${LOG_PREFIX} $(date --utc '+%Y-%m-%dT%H:%M:%SZ') WARN  $*" >&2; }
err()  { echo "${LOG_PREFIX} $(date --utc '+%Y-%m-%dT%H:%M:%SZ') ERROR $*" >&2; }

# ---------------------------------------------------------------------------
# Validate prerequisites
# ---------------------------------------------------------------------------
if [[ ! -f "${CONF_FILE}" ]]; then
    err "Config file not found: ${CONF_FILE}"
    err "Expected format (one line per db): port:database:user:password"
    exit 1
fi

if ! command -v pg_dump &>/dev/null; then
    err "pg_dump not found in PATH. Install postgresql-client."
    exit 1
fi

# ---------------------------------------------------------------------------
# Prepare backup directory
# ---------------------------------------------------------------------------
mkdir -p "${BACKUP_DIR}"
log "Backup directory: ${BACKUP_DIR}"

# ---------------------------------------------------------------------------
# Trap: log any unexpected exit
# ---------------------------------------------------------------------------
trap 'err "Script exited unexpectedly at line ${LINENO}"' ERR

# ---------------------------------------------------------------------------
# Backup loop
# ---------------------------------------------------------------------------
overall_exit=0
success_count=0
fail_count=0

while IFS=: read -r port database user password || [[ -n "${port:-}" ]]; do
    # Skip blank lines and comments
    [[ -z "${port}" || "${port}" == \#* ]] && continue

    dump_file="${BACKUP_DIR}/${database}_${port}.dump"
    log "Dumping ${database} (port ${port}, user ${user}) -> ${dump_file}"

    if PGPASSWORD="${password}" pg_dump \
        --host=localhost \
        --port="${port}" \
        --username="${user}" \
        --format=custom \
        --compress=6 \
        --no-password \
        "${database}" \
        > "${dump_file}" 2>&1; then

        dump_size="$(du -sh "${dump_file}" | cut -f1)"
        log "  OK  ${database} (${dump_size})"
        (( success_count++ )) || true
    else
        err "  FAIL  ${database} (port ${port}) — dump failed"
        # Remove incomplete file to avoid restoring corrupt data
        rm -f "${dump_file}"
        (( fail_count++ )) || true
        overall_exit=1
    fi

done < "${CONF_FILE}"

log "Completed: ${success_count} succeeded, ${fail_count} failed"

# ---------------------------------------------------------------------------
# Prune old backups (keep RETENTION_DAYS worth of directories)
# ---------------------------------------------------------------------------
log "Pruning backup directories older than ${RETENTION_DAYS} days..."
prune_count=0
while IFS= read -r -d '' old_dir; do
    log "  Removing: ${old_dir}"
    rm -rf "${old_dir}"
    (( prune_count++ )) || true
done < <(find "${BACKUP_BASE}" \
    -mindepth 1 \
    -maxdepth 1 \
    -type d \
    -mtime "+${RETENTION_DAYS}" \
    -print0)

log "Pruned ${prune_count} old backup directories"

# ---------------------------------------------------------------------------
# Final status
# ---------------------------------------------------------------------------
if [[ "${overall_exit}" -eq 0 ]]; then
    log "All databases backed up successfully"
else
    err "${fail_count} database(s) failed — review logs above"
fi

exit "${overall_exit}"