lots of stuff

2026-08-16 14:48:04 +02:00
parent c63db540e8
commit 49fa352476
45 changed files with 32520 additions and 93 deletions
@@ -0,0 +1,105 @@
+#!/bin/bash
+
+set -uo pipefail
+
+textfile_dir={{ node_exporter_textfile_dir }}
+file=${textfile_dir}/services.prom
+tmp=${file}.$$
+
+mkdir -p ${textfile_dir}
+
+declare -A active_count
+declare -A unit_count
+units=""
+
+record() {
+    local service=$1
+    local unit=$2
+    local kind=$3
+    local state=$4
+
+    units+="service_unit_active{service=\"${service}\",unit=\"${unit}\",kind=\"${kind}\"} ${state}"$'\n'
+    unit_count[${service}]=$(( ${unit_count[${service}]:-0} + 1 ))
+    active_count[${service}]=$(( ${active_count[${service}]:-0} + state ))
+}
+
+check_user_unit() {
+    local service=$1
+    local owner=$2
+    local unit=$3
+    local state=0
+
+    if systemctl --user --machine ${owner}@.host is-active --quiet ${unit} 2>/dev/null
+    then
+        state=1
+    fi
+
+    record ${service} ${unit} rootless ${state}
+}
+
+check_system_unit() {
+    local service=$1
+    local unit=$2
+    local state=0
+
+    if systemctl is-active --quiet ${unit} 2>/dev/null
+    then
+        state=1
+    fi
+
+    record ${service} ${unit} system ${state}
+}
+
+check_container() {
+    local service=$1
+    local container=$2
+    local state=0
+
+    if [ "$(podman container inspect -f '{% raw %}{{.State.Running}}{% endraw %}' ${container} 2>/dev/null)" = "true" ]
+    then
+        state=1
+    fi
+
+    record ${service} ${container} container ${state}
+}
+
+{% for service in monitored_services %}
+{% for unit in service.units | default([]) %}
+check_user_unit {{ service.name }} {{ service.owner }} {{ unit }}
+{% endfor %}
+{% for unit in service.system_units | default([]) %}
+check_system_unit {{ service.name }} {{ unit }}
+{% endfor %}
+{% for container in service.containers | default([]) %}
+check_container {{ service.name }} {{ container }}
+{% endfor %}
+{% endfor %}
+
+{
+    echo "# HELP service_unit_active Whether a unit or container backing a service is running."
+    echo "# TYPE service_unit_active gauge"
+    printf '%s' "${units}"
+
+    echo "# HELP service_enabled Whether a service is expected to be running."
+    echo "# TYPE service_enabled gauge"
+{% for service in monitored_services %}
+    echo "service_enabled{service=\"{{ service.name }}\",group=\"{{ service.group }}\"} {{ 1 if service.enabled else 0 }}"
+{% endfor %}
+
+    echo "# HELP service_units_total Number of units or containers backing a service."
+    echo "# TYPE service_units_total gauge"
+    for service in "${!unit_count[@]}"
+    do
+        echo "service_units_total{service=\"${service}\"} ${unit_count[${service}]}"
+    done
+
+    echo "# HELP service_units_active Number of running units or containers backing a service."
+    echo "# TYPE service_units_active gauge"
+    for service in "${!active_count[@]}"
+    do
+        echo "service_units_active{service=\"${service}\"} ${active_count[${service}]}"
+    done
+} > "${tmp}"
+
+chmod 0644 "${tmp}"
+mv "${tmp}" "${file}"
@@ -0,0 +1,71 @@
+#!/bin/bash
+
+set -uo pipefail
+
+textfile_dir={{ node_exporter_textfile_dir }}
+file=${textfile_dir}/uptimerobot.prom
+tmp=${file}.$$
+
+mkdir -p ${textfile_dir}
+
+response=$(curl -sS -m 25 -X POST \
+    "https://api.uptimerobot.com/v2/getMonitors?api_key={{ uptimerobot_api_key }}&format=json&custom_uptime_ratios=1-7-30" \
+    2>/dev/null)
+
+{% raw %}
+printf '%s' "${response}" | python3 -c '
+import json
+import sys
+
+print("# HELP uptimerobot_api_ok Whether the uptimerobot api answered successfully.")
+print("# TYPE uptimerobot_api_ok gauge")
+print("# HELP uptimerobot_monitor_up Whether an external monitor reports the target as up.")
+print("# TYPE uptimerobot_monitor_up gauge")
+print("# HELP uptimerobot_monitor_enabled Whether an external monitor is active rather than paused.")
+print("# TYPE uptimerobot_monitor_enabled gauge")
+print("# HELP uptimerobot_monitor_status Raw uptimerobot status, 0 paused 1 unchecked 2 up 8 seems down 9 down.")
+print("# TYPE uptimerobot_monitor_status gauge")
+print("# HELP uptimerobot_uptime_ratio Uptime percentage over the given window.")
+print("# TYPE uptimerobot_uptime_ratio gauge")
+
+try:
+    payload = json.load(sys.stdin)
+except ValueError:
+    print("uptimerobot_api_ok 0")
+    sys.exit(0)
+
+if payload.get("stat") != "ok":
+    print("uptimerobot_api_ok 0")
+    sys.exit(0)
+
+print("uptimerobot_api_ok 1")
+
+def escape(value):
+    return value.replace("\\", "\\\\").replace("\"", "\\\"")
+
+for monitor in payload.get("monitors", []):
+    name = escape(str(monitor.get("friendly_name", "unknown")))
+    status = int(monitor.get("status", 1))
+    labels = "monitor=\"%s\"" % name
+
+    print("uptimerobot_monitor_status{%s} %d" % (labels, status))
+    print("uptimerobot_monitor_enabled{%s} %d" % (labels, 0 if status == 0 else 1))
+    print("uptimerobot_monitor_up{%s} %d" % (labels, 1 if status == 2 else 0))
+
+    ratios = str(monitor.get("custom_uptime_ratio", "")).split("-")
+    for window, ratio in zip(("1d", "7d", "30d"), ratios):
+        try:
+            print("uptimerobot_uptime_ratio{%s,window=\"%s\"} %s" % (labels, window, float(ratio)))
+        except ValueError:
+            pass
+' > "${tmp}"
+{% endraw %}
+
+if [ ! -s "${tmp}" ]
+then
+    rm -f "${tmp}"
+    exit 1
+fi
+
+chmod 0644 "${tmp}"
+mv "${tmp}" "${file}"
@@ -0,0 +1,60 @@
+#!/bin/bash
+
+set -uo pipefail
+
+textfile_dir={{ node_exporter_textfile_dir }}
+file=${textfile_dir}/wireguard.prom
+tmp=${file}.$$
+now=$(date +%s)
+
+mkdir -p ${textfile_dir}
+
+peer_name() {
+    case "$1" in
+{% for key, name in wireguard_peer_names.items() %}
+        "{{ key }}") echo "{{ name }}" ;;
+{% endfor %}
+        *) echo "unknown" ;;
+    esac
+}
+
+{
+    echo "# HELP wireguard_peer_last_handshake_seconds Unix time of the last handshake with a peer."
+    echo "# TYPE wireguard_peer_last_handshake_seconds gauge"
+    echo "# HELP wireguard_peer_handshake_age_seconds Seconds since the last handshake with a peer."
+    echo "# TYPE wireguard_peer_handshake_age_seconds gauge"
+    echo "# HELP wireguard_peer_up Whether a peer handshaked within the last five minutes."
+    echo "# TYPE wireguard_peer_up gauge"
+    echo "# HELP wireguard_peer_receive_bytes_total Bytes received from a peer."
+    echo "# TYPE wireguard_peer_receive_bytes_total counter"
+    echo "# HELP wireguard_peer_transmit_bytes_total Bytes sent to a peer."
+    echo "# TYPE wireguard_peer_transmit_bytes_total counter"
+
+    wg show all dump 2>/dev/null | awk 'NF >= 8' | while read -r interface key psk endpoint allowed handshake rx tx keepalive
+    do
+        name=$(peer_name "${key}")
+        labels="interface=\"${interface}\",peer=\"${name}\""
+
+        age=$((now - handshake))
+        up=0
+
+        if [ "${handshake}" -gt 0 ] && [ "${age}" -lt 300 ]
+        then
+            up=1
+        fi
+
+        if [ "${handshake}" -eq 0 ]
+        then
+            age=-1
+        fi
+
+        echo "wireguard_peer_last_handshake_seconds{${labels}} ${handshake}"
+        echo "wireguard_peer_handshake_age_seconds{${labels}} ${age}"
+        echo "wireguard_peer_up{${labels}} ${up}"
+        echo "wireguard_peer_receive_bytes_total{${labels}} ${rx}"
+        echo "wireguard_peer_transmit_bytes_total{${labels}} ${tx}"
+    done
+} > "${tmp}"
+
+chmod 0644 "${tmp}"
+mv "${tmp}" "${file}"