Files
ProxMenux/scripts/backup_restore/run_scheduled_backup.sh
2026-07-23 22:55:28 +02:00

737 lines
29 KiB
Bash
Executable File

#!/bin/bash
# ==========================================================
# ProxMenux - Run Scheduled Host Backup Job
# ==========================================================
set -u
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
LOCAL_SCRIPTS_LOCAL="$(cd "$SCRIPT_DIR/.." && pwd)"
LOCAL_SCRIPTS_DEFAULT="/usr/local/share/proxmenux/scripts"
LOCAL_SCRIPTS="$LOCAL_SCRIPTS_DEFAULT"
BASE_DIR="/usr/local/share/proxmenux"
UTILS_FILE="$LOCAL_SCRIPTS/utils.sh"
if [[ -f "$LOCAL_SCRIPTS_LOCAL/utils.sh" ]]; then
LOCAL_SCRIPTS="$LOCAL_SCRIPTS_LOCAL"
UTILS_FILE="$LOCAL_SCRIPTS/utils.sh"
elif [[ ! -f "$UTILS_FILE" ]]; then
UTILS_FILE="$BASE_DIR/utils.sh"
fi
if [[ -f "$UTILS_FILE" ]]; then
# shellcheck source=/dev/null
source "$UTILS_FILE"
else
echo "ERROR: utils.sh not found" >&2
exit 1
fi
load_language
initialize_cache
LIB_FILE="$SCRIPT_DIR/lib_host_backup_common.sh"
[[ ! -f "$LIB_FILE" ]] && LIB_FILE="$LOCAL_SCRIPTS_DEFAULT/backup_restore/lib_host_backup_common.sh"
if [[ -f "$LIB_FILE" ]]; then
# shellcheck source=/dev/null
source "$LIB_FILE"
else
echo "ERROR: lib_host_backup_common.sh not found" >&2
exit 1
fi
JOBS_DIR="${PMX_BACKUP_JOBS_DIR:-/var/lib/proxmenux/backup-jobs}"
LOG_DIR="${PMX_BACKUP_LOG_DIR:-/var/log/proxmenux/backup-jobs}"
LOCK_DIR="${PMX_BACKUP_LOCK_DIR:-/var/lock}"
mkdir -p "$JOBS_DIR" "$LOG_DIR" >/dev/null 2>&1 || true
_sb_prune_local() {
local job_id="$1"
local dest_dir="$2"
local ext="$3" # tar.zst or tar.gz
local keep_last="${KEEP_LAST:-0}"
local -a files=()
mapfile -t files < <(find "$dest_dir" -maxdepth 1 -type f -name "${job_id}-*.${ext}" | sort -r)
[[ ${#files[@]} -eq 0 ]] && return 0
if [[ "$keep_last" =~ ^[0-9]+$ ]] && (( keep_last > 0 )); then
local idx=0
for f in "${files[@]}"; do
idx=$((idx+1))
(( idx <= keep_last )) && continue
# Remove the archive AND its sidecar in one shot — if we
# leave .proxmenux.json files behind, the Monitor would
# show them as broken entries pointing at deleted archives.
# Also drop the matching runner log so retention is symmetric:
# one archive ↔ one log, never an orphan log left behind.
local stem suffix
stem=$(basename "$f")
suffix=".${ext}"
stem="${stem%"$suffix"}"
rm -f "$f" "${f}.proxmenux.json" "${LOG_DIR}/${stem}.log" || true
done
fi
}
_sb_run_local() {
local stage_root="$1"
local job_id="$2"
local ts="$3"
local dest_dir="$4"
local archive_ext="${LOCAL_ARCHIVE_EXT:-tar.zst}"
local archive="${dest_dir}/${job_id}-${ts}.${archive_ext}"
mkdir -p "$dest_dir" || return 1
if [[ "$archive_ext" == "tar.zst" ]] && command -v zstd >/dev/null 2>&1; then
tar --zstd -cf "$archive" -C "$stage_root" . >/dev/null 2>&1 || return 1
else
archive="${dest_dir}/${job_id}-${ts}.tar.gz"
tar -czf "$archive" -C "$stage_root" . >/dev/null 2>&1 || return 1
archive_ext="tar.gz"
fi
# Drop a sidecar JSON next to the archive — explicit marker the
# Monitor can use to identify this as a scheduled host backup,
# independent of any future rename of the archive.
hb_write_archive_sidecar "$archive" "scheduled" "$job_id" "${PROFILE:-}" || true
_sb_prune_local "$job_id" "$dest_dir" "$archive_ext"
echo "LOCAL_ARCHIVE=$archive"
return 0
}
# Look up a borg destination's passphrase from the proxmenux state
# directory. The shell convention is `borg-pass-<name>.txt` next to
# `borg-targets.txt`; we resolve the `<name>` by scanning targets for
# the one whose repository matches the runtime BORG_REPO.
_sb_borg_resolve_password() {
local repo="$1"
local cfg="${HB_STATE_DIR:-/usr/local/share/proxmenux}/borg-targets.txt"
[[ -f "$cfg" && -n "$repo" ]] || return 1
local found_name="" line name target_repo
while IFS='|' read -r name target_repo _; do
[[ -z "$name" || -z "$target_repo" ]] && continue
if [[ "$target_repo" == "$repo" ]]; then
found_name="$name"
break
fi
done < "$cfg"
[[ -z "$found_name" ]] && return 1
local pf="${HB_STATE_DIR:-/usr/local/share/proxmenux}/borg-pass-${found_name}.txt"
[[ -r "$pf" ]] || return 1
cat "$pf"
}
_sb_run_borg() {
local stage_root="$1"
local archive_name="$2"
local borg_bin repo passphrase encrypt_mode
borg_bin=$(hb_ensure_borg) || { echo "borg binary not available"; return 1; }
repo="${BORG_REPO:-}"
passphrase="${BORG_PASSPHRASE:-}"
encrypt_mode="${BORG_ENCRYPT_MODE:-none}"
if [[ -z "$repo" ]]; then
echo "BORG_REPO not configured"
return 1
fi
# Empty passphrase + encrypted mode → look it up from the saved
# destination sidecar (the canonical place for borg credentials,
# same pattern as PBS). Jobs created against a destination that
# has its passphrase stored no longer need to duplicate it in the
# job .env.
if [[ -z "$passphrase" && "$encrypt_mode" != "none" ]]; then
passphrase=$(_sb_borg_resolve_password "$repo" 2>/dev/null || true)
fi
if [[ "$encrypt_mode" != "none" && -z "$passphrase" ]]; then
echo "BORG_PASSPHRASE is required when encryption is enabled (mode=$encrypt_mode)"
echo " → save the passphrase on the destination or set it in the job .env"
return 1
fi
# Re-export the credentials so child processes (borg, ssh) inherit
# them. `source <job.env>` only assigns to the current shell — without
# an explicit re-export, child `borg` calls drop back to ssh defaults
# and a remote repo silently auth-fails with no log trail.
export BORG_PASSPHRASE="$passphrase"
[[ -n "${BORG_RSH:-}" ]] && export BORG_RSH
export BORG_RELOCATED_REPO_ACCESS_IS_OK=yes
export BORG_UNKNOWN_UNENCRYPTED_REPO_ACCESS_IS_OK=yes
# Probe the repo first when working with a local path. Three cases
# the operator hits regularly:
# - path is a valid borg repo + creds match → `borg list` works
# - path is a valid borg repo + creds DON'T match → explicit
# "credentials don't match" error (a wrong passphrase saved on
# the destination, typically)
# - path exists, has files, but is NOT a borg repo → explicit
# "path is not empty" error so the operator points at a fresh
# subdirectory instead of hitting borg's confusing "There is
# already something at <path>" message
if [[ "$repo" != ssh://* && -d "$repo" ]]; then
if ! "$borg_bin" list "$repo" </dev/null >/dev/null 2>&1; then
local repo_config="$repo/config"
if [[ -f "$repo_config" ]] && grep -qE '^\[repository\]' "$repo_config" 2>/dev/null; then
echo "Borg repository at ${repo} exists but the configured credentials don't match it."
echo " → the saved passphrase or encryption mode for this destination is wrong."
return 1
fi
# Not a repo. If the directory isn't empty, init will fail with
# "There is already something at <path>". Detect and explain.
if find "$repo" -mindepth 1 -maxdepth 1 -print -quit 2>/dev/null | grep -q .; then
echo "Path ${repo} is not a borg repository AND is not empty."
echo " → Borg refuses to initialize a repo where there are other files."
echo " → Point the destination at a fresh empty subdirectory (e.g. ${repo}/borgbackup)."
return 1
fi
fi
fi
# Send stderr to stdout so the caller's `>>$log_file 2>&1` captures
# borg's diagnostics. The previous `>/dev/null 2>&1` silenced every
# error and made "Job finished with errors" impossible to debug.
if ! hb_borg_init_if_needed "$borg_bin" "$repo" "$encrypt_mode" 2>&1; then
return 1
fi
# Include manifest.json (top-level) when present. Borg needs the
# paths spelled out explicitly — without this, parse_manifest can't
# find the schema'd manifest on a restored Borg archive.
local -a _borg_paths=(rootfs metadata)
[[ -f "$stage_root/manifest.json" ]] && _borg_paths+=(manifest.json)
(cd "$stage_root" && "$borg_bin" create --stats \
"${repo}::${archive_name}" "${_borg_paths[@]}") 2>&1 || return 1
"$borg_bin" prune -v --list "$repo" \
${KEEP_LAST:+--keep-last "$KEEP_LAST"} \
${KEEP_HOURLY:+--keep-hourly "$KEEP_HOURLY"} \
${KEEP_DAILY:+--keep-daily "$KEEP_DAILY"} \
${KEEP_WEEKLY:+--keep-weekly "$KEEP_WEEKLY"} \
${KEEP_MONTHLY:+--keep-monthly "$KEEP_MONTHLY"} \
${KEEP_YEARLY:+--keep-yearly "$KEEP_YEARLY"} \
2>&1 || true
echo "BORG_ARCHIVE=${archive_name}"
return 0
}
# Resolve a PBS password for an auto-discovered Datacenter storage by
# matching the repository string against /etc/pve/storage.cfg and
# returning the contents of /etc/pve/priv/storage/<name>.pw. Used as a
# fallback when the job .env carries an empty PBS_PASSWORD — happens
# when the operator picks a PVE-managed PBS in the wizard (the .pw
# file is the canonical credential location for those).
#
# Implemented in pure bash because the previous awk version relied on
# `match($0, regex, arr)` which is a gawk extension and silently no-ops
# on mawk / busybox awk — leaving the password unresolved and breaking
# every manual / scheduled PBS run against a PVE-managed storage.
_sb_pbs_resolve_password() {
local repo="$1"
local cfg=/etc/pve/storage.cfg
[[ -f "$cfg" && -n "$repo" ]] || return 1
local current_type="" current_name="" server="" datastore="" username=""
local found_name=""
_try_match() {
[[ "$current_type" == "pbs" && -n "$current_name" && -n "$server" && -n "$datastore" ]] || return
local u="${username:-root@pam}"
local built="${u}@${server}:${datastore}"
if [[ "$built" == "$repo" ]]; then
found_name="$current_name"
fi
}
local line key val
while IFS= read -r line; do
if [[ "$line" =~ ^([a-z]+):[[:space:]]+(.+)$ ]]; then
_try_match
[[ -n "$found_name" ]] && break
current_type="${BASH_REMATCH[1]}"
current_name="${BASH_REMATCH[2]}"
server=""; datastore=""; username=""
continue
fi
if [[ "$line" =~ ^[[:space:]]+([a-zA-Z_]+)[[:space:]]+(.+)$ ]]; then
key="${BASH_REMATCH[1]}"
val="${BASH_REMATCH[2]}"
case "$key" in
server) server="$val" ;;
datastore) datastore="$val" ;;
username) username="$val" ;;
esac
fi
done < "$cfg"
# Final block may not have triggered the start-of-next-block check.
[[ -z "$found_name" ]] && _try_match
unset -f _try_match 2>/dev/null
[[ -z "$found_name" ]] && return 1
local pw_file="/etc/pve/priv/storage/${found_name}.pw"
[[ -r "$pw_file" ]] || return 1
cat "$pw_file"
}
# Return the certificate fingerprint of the PBS storage entry whose
# `<user>@<server>:<datastore>` matches $1. Mirrors the walker in
# `_sb_pbs_resolve_password` but reads the `fingerprint <sha>` line
# instead. Needed because attached scheduled jobs are created without
# PBS_FINGERPRINT in their .env (the wizard inherits it from
# `/etc/pve/storage.cfg`), and without a fingerprint the client falls
# back to an interactive `(y/n)` prompt at connect time — which never
# gets an answer under systemd, hanging the whole job. Empty result
# is a valid outcome (PBS behind a CA-signed cert doesn't need one);
# a missing storage entry returns non-zero so the caller can log it.
_sb_pbs_resolve_fingerprint() {
local repo="$1"
local cfg=/etc/pve/storage.cfg
[[ -f "$cfg" && -n "$repo" ]] || return 1
local current_type="" current_name="" server="" datastore="" username=""
local current_fp=""
local found_fp="" found_any=0
local line key val
while IFS= read -r line; do
if [[ "$line" =~ ^([a-z]+):[[:space:]]+(.+)$ ]]; then
if [[ "$current_type" == "pbs" && -n "$current_name" && -n "$server" && -n "$datastore" ]]; then
local u="${username:-root@pam}"
local built="${u}@${server}:${datastore}"
if [[ "$built" == "$repo" ]]; then
found_fp="$current_fp"
found_any=1
break
fi
fi
current_type="${BASH_REMATCH[1]}"
current_name="${BASH_REMATCH[2]}"
server=""; datastore=""; username=""; current_fp=""
continue
fi
if [[ "$line" =~ ^[[:space:]]+([a-zA-Z_]+)[[:space:]]+(.+)$ ]]; then
key="${BASH_REMATCH[1]}"
val="${BASH_REMATCH[2]}"
case "$key" in
server) server="$val" ;;
datastore) datastore="$val" ;;
username) username="$val" ;;
fingerprint) current_fp="$val" ;;
esac
fi
done < "$cfg"
# Final block may not have triggered the start-of-next-block check.
if (( ! found_any )); then
if [[ "$current_type" == "pbs" && -n "$current_name" && -n "$server" && -n "$datastore" ]]; then
local u="${username:-root@pam}"
local built="${u}@${server}:${datastore}"
if [[ "$built" == "$repo" ]]; then
found_fp="$current_fp"
found_any=1
fi
fi
fi
(( found_any )) || return 1
# Empty fingerprint is a legit outcome for CA-signed PBS setups.
printf '%s' "$found_fp"
}
# Same lookup for a PBS destination added through ProxMenux instead of
# Datacenter → Storage. Those never reach storage.cfg, so the walkers
# above find nothing and their credentials live only in our state dir:
# `pbs-manual-configs.txt` holds `<name>|<user>@<server>:<datastore>`
# lines, and the secret itself is in the `pbs-<kind>-<name>.txt` sidecar
# written when the destination was added.
_sb_pbs_state_lookup() {
local repo="$1" kind="$2" # kind: fingerprint | pass
local state_dir="${HB_STATE_DIR:-/usr/local/share/proxmenux}"
local manual="$state_dir/pbs-manual-configs.txt"
[[ -f "$manual" && -n "$repo" ]] || return 1
local line name entry_repo
while IFS= read -r line; do
line="${line%%#*}"
line="${line#"${line%%[![:space:]]*}"}"
line="${line%"${line##*[![:space:]]}"}"
[[ -z "$line" || "$line" != *"|"* ]] && continue
name="${line%%|*}"
entry_repo="${line#*|}"
[[ "$entry_repo" == "$repo" ]] || continue
local f="$state_dir/pbs-${kind}-${name}.txt"
[[ -r "$f" ]] || return 1
tr -d '\r\n' < "$f"
return 0
done < "$manual"
return 1
}
_sb_run_pbs() {
local stage_root="$1"
local backup_id="$2"
local epoch="$3"
# Stage the WHOLE root (rootfs/ + metadata/ + manifest.json), not
# just rootfs/. Mirrors backup_host.sh::_bk_pbs (the TUI flow) — and
# parse_manifest.sh / run_restore.sh need metadata/ + manifest.json
# to compose a meaningful restore plan. With only rootfs/ in the
# pxar, View Contents reports "no manifest.json" forever.
local -a cmd=(
proxmox-backup-client backup
"hostcfg.pxar:${stage_root}"
--repository "$PBS_REPOSITORY"
--backup-type host
--backup-id "$backup_id"
--backup-time "$epoch"
)
# If the .env was created against a Datacenter-managed PBS storage,
# PBS_PASSWORD is intentionally empty there (the credential lives in
# /etc/pve/priv/storage/<name>.pw). Resolve it now so the rest of
# this function can run unchanged.
if [[ -z "${PBS_PASSWORD:-}" && -n "${PBS_REPOSITORY:-}" ]]; then
PBS_PASSWORD=$(_sb_pbs_resolve_password "$PBS_REPOSITORY" 2>/dev/null || true)
# ProxMenux-managed destinations aren't in /etc/pve/priv/storage.
[[ -z "$PBS_PASSWORD" ]] && \
PBS_PASSWORD=$(_sb_pbs_state_lookup "$PBS_REPOSITORY" pass 2>/dev/null || true)
fi
# Same treatment for PBS_FINGERPRINT. Attached scheduled jobs are
# created without one (the wizard delegates trust to PVE's storage
# config), and against a self-signed PBS the client refuses to connect
# without it, so the job fails on every run. Two sources, in order:
# PVE's storage.cfg for Datacenter-managed storages, then our own
# state dir for destinations added from the ProxMenux UI.
if [[ -z "${PBS_FINGERPRINT:-}" && -n "${PBS_REPOSITORY:-}" ]]; then
PBS_FINGERPRINT=$(_sb_pbs_resolve_fingerprint "$PBS_REPOSITORY" 2>/dev/null || true)
[[ -z "$PBS_FINGERPRINT" ]] && \
PBS_FINGERPRINT=$(_sb_pbs_state_lookup "$PBS_REPOSITORY" fingerprint 2>/dev/null || true)
fi
# PBS_ENCRYPTION_PASSWORD: the passphrase that unlocks the keyfile
# itself when it was created with `--kdf scrypt`. The Import flow
# persists it at pbs-key.pass (chmod 600, root-only, same trust
# boundary as the keyfile). If the file is absent the keyfile is
# kdf=none and no passphrase is needed. `PBS_ENCRYPTION_PASSWORD`
# from the .env still wins if the operator set it there for a
# per-job override.
if [[ -z "${PBS_ENCRYPTION_PASSWORD:-}" ]]; then
local _pbs_pass_file="/usr/local/share/proxmenux/pbs-key.pass"
if [[ -r "$_pbs_pass_file" ]]; then
PBS_ENCRYPTION_PASSWORD=$(cat "$_pbs_pass_file" 2>/dev/null || true)
fi
fi
[[ -z "${PBS_REPOSITORY:-}" || -z "${PBS_PASSWORD:-}" ]] && return 1
if [[ -n "${PBS_KEYFILE:-}" ]]; then
cmd+=(--keyfile "$PBS_KEYFILE")
fi
# `</dev/null` on every proxmox-backup-client invocation defends
# against any interactive prompt (fingerprint mismatch, keyfile
# passphrase, ...) that would otherwise block on stdin under systemd
# and freeze the timer for its entire timeout window. With the
# fingerprint resolved above the prompt should never fire, but if
# storage.cfg is out of sync (PBS cert rotated, entry renamed) we
# want a fast, clear failure — not a silent hang.
env PBS_PASSWORD="$PBS_PASSWORD" \
PBS_ENCRYPTION_PASSWORD="${PBS_ENCRYPTION_PASSWORD:-}" \
PBS_FINGERPRINT="${PBS_FINGERPRINT:-}" \
"${cmd[@]}" </dev/null 2>&1 || return 1
# Best effort prune for PBS group.
env PBS_PASSWORD="$PBS_PASSWORD" \
PBS_FINGERPRINT="${PBS_FINGERPRINT:-}" \
proxmox-backup-client prune "host/${backup_id}" --repository "$PBS_REPOSITORY" \
${KEEP_LAST:+--keep-last "$KEEP_LAST"} \
${KEEP_HOURLY:+--keep-hourly "$KEEP_HOURLY"} \
${KEEP_DAILY:+--keep-daily "$KEEP_DAILY"} \
${KEEP_WEEKLY:+--keep-weekly "$KEEP_WEEKLY"} \
${KEEP_MONTHLY:+--keep-monthly "$KEEP_MONTHLY"} \
${KEEP_YEARLY:+--keep-yearly "$KEEP_YEARLY"} \
</dev/null 2>&1 || true
# Upload the keyfile recovery blob alongside the main backup so the
# operator can rebuild the keyfile on a fresh host. Only fires when:
# - this backup was encrypted (PBS_KEYFILE set), AND
# - the operator configured a recovery passphrase
# (pbs-key.recovery.enc exists, created by the UI's
# /pbs-recovery/setup endpoint or shell hb_pbs_setup_recovery).
# The blob is already passphrase-encrypted by openssl, so we do NOT
# pass --keyfile here: PBS stores it as a plain blob retrievable
# without the keyfile, which is the whole point of the escrow.
local recovery_enc="/usr/local/share/proxmenux/pbs-key.recovery.enc"
if [[ -n "${PBS_KEYFILE:-}" && -f "$recovery_enc" ]]; then
local keyrec_id
keyrec_id="hostcfg-$(hostname)-keyrecovery"
env PBS_PASSWORD="$PBS_PASSWORD" \
PBS_FINGERPRINT="${PBS_FINGERPRINT:-}" \
proxmox-backup-client backup \
"keyrecovery.conf:${recovery_enc}" \
--repository "$PBS_REPOSITORY" \
--backup-type host \
--backup-id "$keyrec_id" \
--backup-time "$epoch" \
</dev/null 2>&1 || true
# Prune the paired keyrecovery group with the SAME retention
# values as the main backup. Without this the keyrecovery snapshots
# accumulate one per run while the main group is pruned to keep-*,
# cluttering the datastore over time. Any single keyrecovery snapshot
# is sufficient to reconstruct the keyfile — but matching retention
# keeps the two groups aligned in count and makes cleanup obvious.
env PBS_PASSWORD="$PBS_PASSWORD" \
PBS_FINGERPRINT="${PBS_FINGERPRINT:-}" \
proxmox-backup-client prune "host/${keyrec_id}" --repository "$PBS_REPOSITORY" \
${KEEP_LAST:+--keep-last "$KEEP_LAST"} \
${KEEP_HOURLY:+--keep-hourly "$KEEP_HOURLY"} \
${KEEP_DAILY:+--keep-daily "$KEEP_DAILY"} \
${KEEP_WEEKLY:+--keep-weekly "$KEEP_WEEKLY"} \
${KEEP_MONTHLY:+--keep-monthly "$KEEP_MONTHLY"} \
${KEEP_YEARLY:+--keep-yearly "$KEEP_YEARLY"} \
</dev/null 2>&1 || true
fi
echo "PBS_SNAPSHOT=host/${backup_id}/${epoch}"
return 0
}
# For attached jobs, re-read the parent vzdump job's prune-backups
# live from /etc/pve/jobs.cfg every run and rewrite KEEP_*. The .env
# only carries a snapshot of the parent's retention at creation time,
# so if the parent later gained (or changed) prune-backups the child
# would silently accumulate. Attached mode is authoritative from the
# parent by design — that's the whole point of "attached".
_sb_hydrate_attached_retention() {
local parent="${PVE_PARENT_JOB:-}"
[[ -z "$parent" ]] && return 0
local cfg=/etc/pve/jobs.cfg
[[ -f "$cfg" ]] || return 0
local prune
prune=$(awk -v pid="$parent" '
/^vzdump:[[:space:]]/ { in_block=($2==pid); next }
/^[a-z]+:/ { in_block=0; next }
in_block && /^[[:space:]]+prune-backups[[:space:]]/ {
sub(/^[[:space:]]+prune-backups[[:space:]]+/, "");
print; exit
}
' "$cfg")
# Clear any KEEP_* the .env may have — those are the create-time
# snapshot and are now stale. If the parent later removed prune-
# backups altogether, this correctly stops pruning the child too.
unset KEEP_LAST KEEP_HOURLY KEEP_DAILY KEEP_WEEKLY KEEP_MONTHLY KEEP_YEARLY
[[ -z "$prune" ]] && return 0
local kv
while IFS= read -r kv; do
[[ -n "$kv" ]] && export "${kv?}"
done < <(hb_pve_prune_to_keep_env "$prune")
}
main() {
local job_id="${1:-}"
[[ -z "$job_id" ]] && { echo "Usage: $0 <job_id>" >&2; exit 1; }
local job_file="${JOBS_DIR}/${job_id}.env"
[[ -f "$job_file" ]] || { echo "Job not found: $job_id" >&2; exit 1; }
# shellcheck source=/dev/null
source "$job_file"
# Attached jobs: re-read retention from the PVE parent live (see
# _sb_hydrate_attached_retention above for the why). Standalone
# jobs keep whatever KEEP_* the .env has.
_sb_hydrate_attached_retention
local lock_file="${LOCK_DIR}/proxmenux-backup-${job_id}.lock"
if command -v flock >/dev/null 2>&1; then
exec 9>"$lock_file" || exit 1
if ! flock -n 9; then
echo "Another run is active for job ${job_id}" >&2
exit 1
fi
fi
local ts log_file stage_root summary_file
ts="$(date +%Y%m%d_%H%M%S)"
log_file="${LOG_DIR}/${job_id}-${ts}.log"
summary_file="${LOG_DIR}/${job_id}-last.status"
stage_root="$(mktemp -d /tmp/proxmenux-sched-stage.XXXXXX)"
{
echo "JOB_ID=${job_id}"
echo "RUN_AT=$(date -Iseconds)"
echo "BACKEND=${BACKEND:-}"
echo "PROFILE_MODE=${PROFILE_MODE:-default}"
} >"$summary_file"
{
echo "=== Scheduled backup job ${job_id} started at $(date -Iseconds) ==="
echo "Backend: ${BACKEND:-}"
echo "Profile: ${PROFILE_MODE:-default}"
} >"$log_file"
# Fire the lifecycle "start" event. Per-channel/per-event toggles in
# Settings decide whether the user actually receives it
# (host_backup_start ships with default_enabled=False).
local _hb_dest=""
case "${BACKEND:-}" in
local) _hb_dest="${LOCAL_DEST_DIR:-/var/lib/vz/dump}" ;;
borg) _hb_dest="${BORG_REPO:-}" ;;
pbs) _hb_dest="${PBS_REPOSITORY:-}" ;;
esac
export HB_NOTIFY_BACKEND="${BACKEND:-}"
export HB_NOTIFY_DESTINATION="$_hb_dest"
export HB_NOTIFY_PROFILE_MODE="${PROFILE_MODE:-default}"
export HB_NOTIFY_LOG_FILE="$log_file"
hb_notify_lifecycle "start"
local -a paths=()
if [[ "${PROFILE_MODE:-default}" == "custom" && -f "${JOBS_DIR}/${job_id}.paths" ]]; then
mapfile -t paths < "${JOBS_DIR}/${job_id}.paths"
else
# Default profile = base paths + operator-saved extras (the
# TUI flow does the same — see hb_resolve_paths_mode in
# lib_host_backup_common.sh). Without the extras line, the
# `backup-extra-paths.txt` set in Settings was silently
# ignored for scheduled / manual runs from the Monitor.
mapfile -t paths < <(hb_default_profile_paths; hb_load_extra_paths)
fi
if [[ ${#paths[@]} -eq 0 ]]; then
echo "No paths configured for job" >>"$log_file"
echo "RESULT=failed" >>"$summary_file"
rm -rf "$stage_root"
exit 1
fi
# Interactive output mirrors the colored layout of _bk_local in
# backup_host.sh when stdout is a TTY (operator launched "Run job
# now"). Otherwise — timer / vzdump hook — only the plain log
# file is written.
local TTY=0
[[ -t 1 ]] && TTY=1
if (( TTY )); then
echo -e "${TAB}${BGN}$(translate "Backend:")${CL} ${BL}${BACKEND}${CL}"
echo -e "${TAB}${BGN}$(translate "Profile:")${CL} ${BL}${PROFILE_MODE:-default}${CL}"
echo -e "${TAB}${BGN}$(translate "Paths to back up:")${CL} ${BL}${#paths[@]}${CL}"
echo
msg_info "$(translate "Preparing staging area...")"
fi
{
echo "Paths to back up: ${#paths[@]}"
echo "Preparing staging area at $stage_root ..."
} >>"$log_file"
hb_prepare_staging "$stage_root" "${paths[@]}" >>"$log_file" 2>&1
local staged_files staged_size
staged_files=$(find "$stage_root/rootfs" -type f 2>/dev/null | wc -l)
staged_size=$(hb_file_size "$stage_root/rootfs" 2>/dev/null || echo "?")
echo "Staging ready: $staged_files files copied (size $staged_size)." >>"$log_file"
(( TTY )) && msg_ok "$(translate "Staging ready.") $(translate "Data size:") $staged_size$staged_files $(translate "files")"
local rc=1 t_start elapsed archive_path=""
t_start=$SECONDS
case "${BACKEND:-}" in
local)
(( TTY )) && { echo; msg_info "$(translate "Creating local archive...")"; stop_spinner; }
echo "Writing local archive to ${LOCAL_DEST_DIR:-/var/lib/vz/dump} ..." >>"$log_file"
local _output
_output=$(_sb_run_local "$stage_root" "$job_id" "$ts" "${LOCAL_DEST_DIR:-/var/lib/vz/dump}" 2>>"$log_file")
rc=$?
echo "$_output" >>"$log_file"
archive_path=$(grep "^LOCAL_ARCHIVE=" <<<"$_output" | cut -d'=' -f2-)
;;
borg)
(( TTY )) && { echo; msg_info "$(translate "Sending backup to Borg repository...")"; stop_spinner; }
echo "Sending backup to Borg repository ${BORG_REPO:-} ..." >>"$log_file"
_sb_run_borg "$stage_root" "${job_id}-${ts}" >>"$log_file" 2>&1
rc=$?
archive_path="${BORG_REPO:-}::${job_id}-${ts}"
;;
pbs)
(( TTY )) && { echo; msg_info "$(translate "Sending backup to PBS...")"; stop_spinner; }
echo "Sending backup to PBS ${PBS_REPOSITORY:-} (id=${PBS_BACKUP_ID:-hostcfg-$(hostname)}) ..." >>"$log_file"
_sb_run_pbs "$stage_root" "${PBS_BACKUP_ID:-hostcfg-$(hostname)}" "$(date +%s)" >>"$log_file" 2>&1
rc=$?
archive_path="${PBS_REPOSITORY:-}::host/${PBS_BACKUP_ID:-hostcfg-$(hostname)}"
;;
*)
echo "Unknown backend: ${BACKEND:-}" >>"$log_file"
rc=1
;;
esac
elapsed=$((SECONDS - t_start))
echo "Cleaning up staging area ..." >>"$log_file"
rm -rf "$stage_root"
# Single source of truth for archive_size (was previously computed
# only inside the TTY block, which left the notification body with
# '-' for headless scheduled runs). Local backends know the actual
# output file; PBS / Borg backends manage their own dedupe storage
# so we leave '-' meaning "see repository for usage".
local archive_size="-"
if [[ "${BACKEND:-}" == "local" && -n "$archive_path" && -f "$archive_path" ]]; then
archive_size=$(hb_file_size "$archive_path" 2>/dev/null || echo '-')
fi
local pretty_duration
pretty_duration=$(hb_human_elapsed "$elapsed" 2>/dev/null || echo "${elapsed}s")
if [[ $rc -eq 0 ]]; then
echo "RESULT=ok" >>"$summary_file"
echo "LOG_FILE=${log_file}" >>"$summary_file"
echo "=== Job finished OK at $(date -Iseconds) ===" >>"$log_file"
export HB_NOTIFY_DATA_SIZE="$staged_size"
export HB_NOTIFY_ARCHIVE_SIZE="$archive_size"
export HB_NOTIFY_DURATION="$pretty_duration"
hb_notify_lifecycle "complete"
if (( TTY )); then
local method_label
case "${BACKEND:-}" in
local) method_label="Local archive (tar)" ;;
borg) method_label="Borg repository" ;;
pbs) method_label="Proxmox Backup Server" ;;
esac
echo
echo -e "${TAB}${BOLD}$(translate "Backup completed:")${CL}"
echo -e "${TAB}${BGN}$(translate "Method:")${CL} ${BL}${method_label}${CL}"
[[ -n "$archive_path" ]] && \
echo -e "${TAB}${BGN}$(translate "Archive:")${CL} ${BL}${archive_path}${CL}"
echo -e "${TAB}${BGN}$(translate "Data size:")${CL} ${BL}${staged_size}${CL}"
[[ "$archive_size" != "-" ]] && \
echo -e "${TAB}${BGN}$(translate "Archive size:")${CL} ${BL}${archive_size}${CL}"
echo -e "${TAB}${BGN}$(translate "Duration:")${CL} ${BL}$(hb_human_elapsed "$elapsed")${CL}"
echo -e "${TAB}${BGN}$(translate "Log:")${CL} ${BL}${log_file}${CL}"
echo
msg_ok "$(translate "Backup completed successfully.")"
fi
exit 0
else
echo "RESULT=failed" >>"$summary_file"
echo "LOG_FILE=${log_file}" >>"$summary_file"
echo "=== Job finished with errors at $(date -Iseconds) ===" >>"$log_file"
# Build a one-line reason from the last error-ish line in the log
# so the notification body has something actionable rather than a
# generic "Reason: unknown". Falls back to the literal exit code.
local _hb_reason=""
_hb_reason=$(grep -iE 'error|fail|fatal|abort' "$log_file" 2>/dev/null | tail -1 | sed 's/^[[:space:]]*//')
[[ -z "$_hb_reason" ]] && _hb_reason="Runner exited with status ${rc}"
export HB_NOTIFY_DATA_SIZE="${staged_size:-}"
export HB_NOTIFY_ARCHIVE_SIZE="$archive_size"
export HB_NOTIFY_DURATION="$pretty_duration"
export HB_NOTIFY_REASON="$_hb_reason"
hb_notify_lifecycle "fail"
(( TTY )) && msg_error "$(translate "Backup failed. See log:") $log_file"
exit 1
fi
}
main "$@"