feat: add CUPS watchdog timers
This commit is contained in:
8
watchdog/cups-network-watchdog.service
Normal file
8
watchdog/cups-network-watchdog.service
Normal file
@@ -0,0 +1,8 @@
|
||||
[Unit]
|
||||
Description=CUPS network watchdog
|
||||
After=network-online.target
|
||||
Wants=network-online.target
|
||||
|
||||
[Service]
|
||||
Type=oneshot
|
||||
ExecStart=/opt/cups-watchdog/network-watchdog.sh check
|
||||
11
watchdog/cups-network-watchdog.timer
Normal file
11
watchdog/cups-network-watchdog.timer
Normal file
@@ -0,0 +1,11 @@
|
||||
[Unit]
|
||||
Description=Run CUPS network watchdog every minute
|
||||
|
||||
[Timer]
|
||||
OnBootSec=2min
|
||||
OnUnitActiveSec=1min
|
||||
AccuracySec=15s
|
||||
Persistent=true
|
||||
|
||||
[Install]
|
||||
WantedBy=timers.target
|
||||
8
watchdog/cups-print-watchdog.service
Normal file
8
watchdog/cups-print-watchdog.service
Normal file
@@ -0,0 +1,8 @@
|
||||
[Unit]
|
||||
Description=CUPS print service watchdog
|
||||
After=cups.service
|
||||
Wants=cups.service
|
||||
|
||||
[Service]
|
||||
Type=oneshot
|
||||
ExecStart=/opt/cups-watchdog/print-watchdog.sh check
|
||||
11
watchdog/cups-print-watchdog.timer
Normal file
11
watchdog/cups-print-watchdog.timer
Normal file
@@ -0,0 +1,11 @@
|
||||
[Unit]
|
||||
Description=Run CUPS print service watchdog every minute
|
||||
|
||||
[Timer]
|
||||
OnBootSec=2min
|
||||
OnUnitActiveSec=1min
|
||||
AccuracySec=15s
|
||||
Persistent=true
|
||||
|
||||
[Install]
|
||||
WantedBy=timers.target
|
||||
275
watchdog/network-watchdog.sh
Normal file
275
watchdog/network-watchdog.sh
Normal file
@@ -0,0 +1,275 @@
|
||||
#!/bin/bash
|
||||
#
|
||||
# Network watchdog for the CUPS print server.
|
||||
# Default behavior: when connectivity checks fail repeatedly, switch the
|
||||
# configured interface back to DHCP so the box can regain network access.
|
||||
#
|
||||
# Edit /etc/cups-watchdog/network-watchdog.conf to set STATIC_IP,
|
||||
# STATIC_PREFIX, STATIC_GATEWAY and STATIC_DNS. You can also run:
|
||||
# /opt/cups-watchdog/network-watchdog.sh static
|
||||
# /opt/cups-watchdog/network-watchdog.sh dhcp
|
||||
#
|
||||
|
||||
set -u
|
||||
|
||||
CONFIG_FILE="${CUPS_NETWORK_WATCHDOG_CONFIG:-/etc/cups-watchdog/network-watchdog.conf}"
|
||||
STATE_DIR="/run/cups-watchdog"
|
||||
|
||||
INTERFACE=""
|
||||
STATIC_IP=""
|
||||
STATIC_PREFIX="24"
|
||||
STATIC_GATEWAY=""
|
||||
STATIC_DNS="114.114.114.114 223.5.5.5"
|
||||
PING_TARGETS="223.5.5.5 114.114.114.114"
|
||||
FAIL_THRESHOLD=3
|
||||
DHCP_AFTER_FAILURE=1
|
||||
LOG_FILE="/var/log/cups-watchdog/network.log"
|
||||
|
||||
if [ -f "$CONFIG_FILE" ]; then
|
||||
# shellcheck disable=SC1090
|
||||
. "$CONFIG_FILE"
|
||||
fi
|
||||
|
||||
mkdir -p "$STATE_DIR"
|
||||
mkdir -p "$(dirname "$LOG_FILE")" 2>/dev/null || true
|
||||
|
||||
log_msg() {
|
||||
local msg="$1"
|
||||
local line
|
||||
line="$(date '+%Y-%m-%d %H:%M:%S') [network-watchdog] $msg"
|
||||
echo "$line"
|
||||
echo "$line" >> "$LOG_FILE" 2>/dev/null || true
|
||||
}
|
||||
|
||||
detect_interface() {
|
||||
if [ -n "${INTERFACE:-}" ] && ip link show "$INTERFACE" >/dev/null 2>&1; then
|
||||
echo "$INTERFACE"
|
||||
return 0
|
||||
fi
|
||||
|
||||
ip route show default 2>/dev/null | awk '{print $5; exit}'
|
||||
}
|
||||
|
||||
nm_connection_for_interface() {
|
||||
local iface="$1"
|
||||
nmcli -t -f NAME,DEVICE con show --active 2>/dev/null | awk -F: -v iface="$iface" '$2 == iface {print $1; exit}'
|
||||
}
|
||||
|
||||
configure_dhcp_nmcli() {
|
||||
local iface="$1"
|
||||
local conn
|
||||
conn="$(nm_connection_for_interface "$iface")"
|
||||
[ -n "$conn" ] || conn="$iface"
|
||||
|
||||
nmcli con mod "$conn" ipv4.method auto ipv4.addresses "" ipv4.gateway "" ipv4.dns "" >/dev/null
|
||||
nmcli con down "$conn" >/dev/null 2>&1 || true
|
||||
nmcli con up "$conn" >/dev/null
|
||||
}
|
||||
|
||||
configure_dhcp_netplan() {
|
||||
local iface="$1"
|
||||
local target_file="/etc/netplan/99-cups-watchdog-dhcp.yaml"
|
||||
|
||||
cat > "$target_file" << EOF
|
||||
# DHCP recovery config generated by cups network watchdog.
|
||||
network:
|
||||
version: 2
|
||||
renderer: networkd
|
||||
ethernets:
|
||||
$iface:
|
||||
dhcp4: yes
|
||||
EOF
|
||||
chmod 600 "$target_file"
|
||||
rm -f /etc/netplan/01-static-ip.yaml /etc/netplan/99-cups-static-ip.yaml /etc/netplan/99-cups-watchdog-static.yaml 2>/dev/null || true
|
||||
netplan apply
|
||||
}
|
||||
|
||||
configure_dhcp_interfaces() {
|
||||
local iface="$1"
|
||||
|
||||
if [ -f /etc/network/interfaces ]; then
|
||||
cp /etc/network/interfaces /etc/network/interfaces.cups-watchdog.bak
|
||||
fi
|
||||
|
||||
cat > /etc/network/interfaces << EOF
|
||||
# DHCP recovery config generated by cups network watchdog.
|
||||
auto lo
|
||||
iface lo inet loopback
|
||||
|
||||
auto $iface
|
||||
iface $iface inet dhcp
|
||||
EOF
|
||||
systemctl restart networking 2>/dev/null || /etc/init.d/networking restart 2>/dev/null || true
|
||||
}
|
||||
|
||||
configure_dhcp() {
|
||||
local iface
|
||||
iface="$(detect_interface)"
|
||||
if [ -z "$iface" ]; then
|
||||
log_msg "cannot switch to DHCP: network interface not found"
|
||||
return 1
|
||||
fi
|
||||
|
||||
log_msg "switching $iface to DHCP"
|
||||
if command -v nmcli >/dev/null 2>&1 && systemctl is-active --quiet NetworkManager 2>/dev/null; then
|
||||
configure_dhcp_nmcli "$iface"
|
||||
elif [ -d /etc/netplan ] && command -v netplan >/dev/null 2>&1; then
|
||||
configure_dhcp_netplan "$iface"
|
||||
else
|
||||
configure_dhcp_interfaces "$iface"
|
||||
fi
|
||||
}
|
||||
|
||||
configure_static_nmcli() {
|
||||
local iface="$1"
|
||||
local conn
|
||||
conn="$(nm_connection_for_interface "$iface")"
|
||||
[ -n "$conn" ] || conn="$iface"
|
||||
|
||||
nmcli con mod "$conn" ipv4.method manual ipv4.addresses "$STATIC_IP/$STATIC_PREFIX" ipv4.gateway "$STATIC_GATEWAY" ipv4.dns "$STATIC_DNS" >/dev/null
|
||||
nmcli con down "$conn" >/dev/null 2>&1 || true
|
||||
nmcli con up "$conn" >/dev/null
|
||||
}
|
||||
|
||||
configure_static_netplan() {
|
||||
local iface="$1"
|
||||
local target_file="/etc/netplan/99-cups-watchdog-static.yaml"
|
||||
local dns_list=""
|
||||
local d
|
||||
|
||||
for d in $STATIC_DNS; do
|
||||
dns_list="$dns_list, $d"
|
||||
done
|
||||
dns_list="${dns_list#, }"
|
||||
|
||||
cat > "$target_file" << EOF
|
||||
# Static network config generated by cups network watchdog.
|
||||
network:
|
||||
version: 2
|
||||
renderer: networkd
|
||||
ethernets:
|
||||
$iface:
|
||||
dhcp4: no
|
||||
addresses:
|
||||
- $STATIC_IP/$STATIC_PREFIX
|
||||
routes:
|
||||
- to: default
|
||||
via: $STATIC_GATEWAY
|
||||
nameservers:
|
||||
addresses: [$dns_list]
|
||||
EOF
|
||||
chmod 600 "$target_file"
|
||||
rm -f /etc/netplan/01-dhcp.yaml /etc/netplan/99-cups-dhcp.yaml /etc/netplan/99-cups-watchdog-dhcp.yaml 2>/dev/null || true
|
||||
netplan apply
|
||||
}
|
||||
|
||||
configure_static_interfaces() {
|
||||
local iface="$1"
|
||||
local netmask="255.255.255.0"
|
||||
|
||||
case "$STATIC_PREFIX" in
|
||||
8) netmask="255.0.0.0" ;;
|
||||
16) netmask="255.255.0.0" ;;
|
||||
24) netmask="255.255.255.0" ;;
|
||||
25) netmask="255.255.255.128" ;;
|
||||
26) netmask="255.255.255.192" ;;
|
||||
27) netmask="255.255.255.224" ;;
|
||||
28) netmask="255.255.255.240" ;;
|
||||
29) netmask="255.255.255.248" ;;
|
||||
30) netmask="255.255.255.252" ;;
|
||||
esac
|
||||
|
||||
if [ -f /etc/network/interfaces ]; then
|
||||
cp /etc/network/interfaces /etc/network/interfaces.cups-watchdog.bak
|
||||
fi
|
||||
|
||||
cat > /etc/network/interfaces << EOF
|
||||
# Static network config generated by cups network watchdog.
|
||||
auto lo
|
||||
iface lo inet loopback
|
||||
|
||||
auto $iface
|
||||
iface $iface inet static
|
||||
address $STATIC_IP
|
||||
netmask $netmask
|
||||
gateway $STATIC_GATEWAY
|
||||
dns-nameservers $STATIC_DNS
|
||||
EOF
|
||||
systemctl restart networking 2>/dev/null || /etc/init.d/networking restart 2>/dev/null || true
|
||||
}
|
||||
|
||||
configure_static() {
|
||||
local iface
|
||||
iface="$(detect_interface)"
|
||||
|
||||
if [ -z "$iface" ]; then
|
||||
log_msg "cannot switch to static: network interface not found"
|
||||
return 1
|
||||
fi
|
||||
|
||||
if [ -z "${STATIC_IP:-}" ] || [ -z "${STATIC_GATEWAY:-}" ]; then
|
||||
log_msg "cannot switch to static: STATIC_IP or STATIC_GATEWAY is empty in $CONFIG_FILE"
|
||||
return 1
|
||||
fi
|
||||
|
||||
log_msg "switching $iface to static $STATIC_IP/$STATIC_PREFIX"
|
||||
if command -v nmcli >/dev/null 2>&1 && systemctl is-active --quiet NetworkManager 2>/dev/null; then
|
||||
configure_static_nmcli "$iface"
|
||||
elif [ -d /etc/netplan ] && command -v netplan >/dev/null 2>&1; then
|
||||
configure_static_netplan "$iface"
|
||||
else
|
||||
configure_static_interfaces "$iface"
|
||||
fi
|
||||
}
|
||||
|
||||
connectivity_ok() {
|
||||
local target
|
||||
|
||||
for target in $PING_TARGETS; do
|
||||
[ -n "$target" ] || continue
|
||||
if ping -c 1 -W 2 "$target" >/dev/null 2>&1; then
|
||||
return 0
|
||||
fi
|
||||
done
|
||||
|
||||
return 1
|
||||
}
|
||||
|
||||
check_network() {
|
||||
local count_file="$STATE_DIR/network.fail_count"
|
||||
local fail_count=0
|
||||
|
||||
if connectivity_ok; then
|
||||
echo 0 > "$count_file"
|
||||
log_msg "connectivity ok"
|
||||
return 0
|
||||
fi
|
||||
|
||||
if [ -f "$count_file" ]; then
|
||||
fail_count="$(cat "$count_file" 2>/dev/null || echo 0)"
|
||||
fi
|
||||
fail_count=$((fail_count + 1))
|
||||
echo "$fail_count" > "$count_file"
|
||||
|
||||
log_msg "connectivity failed ($fail_count/$FAIL_THRESHOLD)"
|
||||
if [ "$fail_count" -ge "$FAIL_THRESHOLD" ] && [ "$DHCP_AFTER_FAILURE" = "1" ]; then
|
||||
configure_dhcp
|
||||
echo 0 > "$count_file"
|
||||
fi
|
||||
}
|
||||
|
||||
case "${1:-check}" in
|
||||
check)
|
||||
check_network
|
||||
;;
|
||||
dhcp)
|
||||
configure_dhcp
|
||||
;;
|
||||
static)
|
||||
configure_static
|
||||
;;
|
||||
*)
|
||||
echo "Usage: $0 [check|dhcp|static]"
|
||||
exit 2
|
||||
;;
|
||||
esac
|
||||
162
watchdog/print-watchdog.sh
Normal file
162
watchdog/print-watchdog.sh
Normal file
@@ -0,0 +1,162 @@
|
||||
#!/bin/bash
|
||||
#
|
||||
# Print service watchdog for the CUPS print server.
|
||||
# It checks systemd service state plus quick CUPS command/HTTP probes.
|
||||
# When checks fail repeatedly, it restarts CUPS and related services.
|
||||
#
|
||||
|
||||
set -u
|
||||
|
||||
CONFIG_FILE="${CUPS_PRINT_WATCHDOG_CONFIG:-/etc/cups-watchdog/print-watchdog.conf}"
|
||||
STATE_DIR="/run/cups-watchdog"
|
||||
|
||||
SERVICES="cups avahi-daemon cups-driver-manager"
|
||||
CUPS_URL="http://127.0.0.1:631/"
|
||||
CHECK_CUPS_HTTP=1
|
||||
CHECK_LPSTAT=1
|
||||
COMMAND_TIMEOUT=8
|
||||
FAIL_THRESHOLD=2
|
||||
RESTART_COOLDOWN=60
|
||||
LOG_FILE="/var/log/cups-watchdog/print.log"
|
||||
|
||||
if [ -f "$CONFIG_FILE" ]; then
|
||||
# shellcheck disable=SC1090
|
||||
. "$CONFIG_FILE"
|
||||
fi
|
||||
|
||||
mkdir -p "$STATE_DIR"
|
||||
mkdir -p "$(dirname "$LOG_FILE")" 2>/dev/null || true
|
||||
|
||||
log_msg() {
|
||||
local msg="$1"
|
||||
local line
|
||||
line="$(date '+%Y-%m-%d %H:%M:%S') [print-watchdog] $msg"
|
||||
echo "$line"
|
||||
echo "$line" >> "$LOG_FILE" 2>/dev/null || true
|
||||
}
|
||||
|
||||
unit_exists() {
|
||||
local service="$1"
|
||||
|
||||
systemctl list-unit-files --no-legend "$service.service" 2>/dev/null | awk '{print $1}' | grep -qx "$service.service" && return 0
|
||||
systemctl list-units --all --no-legend "$service.service" 2>/dev/null | awk '{print $1}' | grep -qx "$service.service"
|
||||
}
|
||||
|
||||
service_should_check() {
|
||||
local service="$1"
|
||||
|
||||
unit_exists "$service" || return 1
|
||||
if systemctl is-enabled --quiet "$service.service" 2>/dev/null; then
|
||||
return 0
|
||||
fi
|
||||
if systemctl is-active --quiet "$service.service" 2>/dev/null; then
|
||||
return 0
|
||||
fi
|
||||
|
||||
return 1
|
||||
}
|
||||
|
||||
check_systemd_services() {
|
||||
local service
|
||||
local failed=0
|
||||
|
||||
for service in $SERVICES; do
|
||||
service_should_check "$service" || continue
|
||||
if ! systemctl is-active --quiet "$service.service" 2>/dev/null; then
|
||||
log_msg "service not active: $service"
|
||||
failed=1
|
||||
fi
|
||||
done
|
||||
|
||||
return "$failed"
|
||||
}
|
||||
|
||||
check_lpstat() {
|
||||
[ "$CHECK_LPSTAT" = "1" ] || return 0
|
||||
command -v lpstat >/dev/null 2>&1 || return 0
|
||||
|
||||
timeout "$COMMAND_TIMEOUT" lpstat -r >/dev/null 2>&1
|
||||
}
|
||||
|
||||
check_http() {
|
||||
[ "$CHECK_CUPS_HTTP" = "1" ] || return 0
|
||||
command -v curl >/dev/null 2>&1 || return 0
|
||||
|
||||
curl -fsS --max-time "$COMMAND_TIMEOUT" "$CUPS_URL" >/dev/null 2>&1
|
||||
}
|
||||
|
||||
restart_print_stack() {
|
||||
local now
|
||||
local last_file="$STATE_DIR/print.last_restart"
|
||||
local last=0
|
||||
local service
|
||||
|
||||
now="$(date +%s)"
|
||||
if [ -f "$last_file" ]; then
|
||||
last="$(cat "$last_file" 2>/dev/null || echo 0)"
|
||||
fi
|
||||
|
||||
if [ $((now - last)) -lt "$RESTART_COOLDOWN" ]; then
|
||||
log_msg "restart skipped: cooldown active"
|
||||
return 0
|
||||
fi
|
||||
|
||||
log_msg "restarting print stack"
|
||||
systemctl reset-failed cups.service >/dev/null 2>&1 || true
|
||||
systemctl restart cups.service
|
||||
|
||||
for service in avahi-daemon cups-driver-manager; do
|
||||
service_should_check "$service" || continue
|
||||
systemctl reset-failed "$service.service" >/dev/null 2>&1 || true
|
||||
systemctl restart "$service.service" >/dev/null 2>&1 || true
|
||||
done
|
||||
|
||||
echo "$now" > "$last_file"
|
||||
}
|
||||
|
||||
check_print_stack() {
|
||||
local count_file="$STATE_DIR/print.fail_count"
|
||||
local fail_count=0
|
||||
local failed=0
|
||||
|
||||
check_systemd_services || failed=1
|
||||
if ! check_lpstat; then
|
||||
log_msg "lpstat check failed or timed out"
|
||||
failed=1
|
||||
fi
|
||||
if ! check_http; then
|
||||
log_msg "CUPS HTTP check failed or timed out"
|
||||
failed=1
|
||||
fi
|
||||
|
||||
if [ "$failed" = "0" ]; then
|
||||
echo 0 > "$count_file"
|
||||
log_msg "print stack ok"
|
||||
return 0
|
||||
fi
|
||||
|
||||
if [ -f "$count_file" ]; then
|
||||
fail_count="$(cat "$count_file" 2>/dev/null || echo 0)"
|
||||
fi
|
||||
fail_count=$((fail_count + 1))
|
||||
echo "$fail_count" > "$count_file"
|
||||
|
||||
log_msg "print stack failed ($fail_count/$FAIL_THRESHOLD)"
|
||||
if [ "$fail_count" -ge "$FAIL_THRESHOLD" ]; then
|
||||
restart_print_stack
|
||||
echo 0 > "$count_file"
|
||||
fi
|
||||
}
|
||||
|
||||
case "${1:-check}" in
|
||||
check)
|
||||
check_print_stack
|
||||
;;
|
||||
restart)
|
||||
restart_print_stack
|
||||
;;
|
||||
*)
|
||||
echo "Usage: $0 [check|restart]"
|
||||
exit 2
|
||||
;;
|
||||
esac
|
||||
Reference in New Issue
Block a user