|
| 1 | +#!/usr/bin/env python3 |
| 2 | + |
| 3 | +""" |
| 4 | +Description: Expose metrics from needrestart. |
| 5 | +
|
| 6 | +This script runs needrestart in batch mode. It will never ask for input |
| 7 | +and will never restart or upgrade anything. |
| 8 | +
|
| 9 | +Dependencies: python >= 3.5, python3-prometheus-client, needrestart |
| 10 | +
|
| 11 | +Authors: RomainMou |
| 12 | +""" |
| 13 | + |
| 14 | +import sys |
| 15 | +import time |
| 16 | +import subprocess |
| 17 | +from collections import Counter |
| 18 | +from enum import Enum |
| 19 | + |
| 20 | +from prometheus_client import ( |
| 21 | + CollectorRegistry, |
| 22 | + Gauge, |
| 23 | + generate_latest, |
| 24 | +) |
| 25 | + |
| 26 | + |
| 27 | +class KernelStatus(Enum): |
| 28 | + UNKNOWN = 0 |
| 29 | + CURRENT = 1 |
| 30 | + ABI_UPGRADE = 2 |
| 31 | + VERSION_UPGRADE = 3 |
| 32 | + |
| 33 | + |
| 34 | +class MicroCodeStatus(Enum): |
| 35 | + UNKNOWN = 0 |
| 36 | + CURRENT = 1 |
| 37 | + OBSOLETE = 2 |
| 38 | + |
| 39 | + |
| 40 | +class NeedRestartData: |
| 41 | + def __init__(self, needrestart_output): |
| 42 | + # Some default value |
| 43 | + self.timestamp = int(time.time()) |
| 44 | + self.version = None |
| 45 | + self.kernel_status = None |
| 46 | + self.microcode_status = None |
| 47 | + self.kernel_current_version = "" |
| 48 | + self.kernel_expected_version = "" |
| 49 | + self.microcode_current_version = "" |
| 50 | + self.microcode_expected_version = "" |
| 51 | + needrestart_counter = Counter() |
| 52 | + |
| 53 | + # Parse the cmd output |
| 54 | + for line in needrestart_output.splitlines(): |
| 55 | + key, value = line.split(": ", maxsplit=1) |
| 56 | + if key == "NEEDRESTART-VER": |
| 57 | + self.version = value |
| 58 | + # Kernel informations |
| 59 | + elif key == "NEEDRESTART-KCUR": |
| 60 | + self.kernel_current_version = value |
| 61 | + elif key == "NEEDRESTART-KEXP": |
| 62 | + self.kernel_expected_version = value |
| 63 | + elif key == "NEEDRESTART-KSTA": |
| 64 | + self.kernel_status = KernelStatus(int(value)) |
| 65 | + # Microcode informations |
| 66 | + elif key == "NEEDRESTART-UCCUR": |
| 67 | + self.microcode_current_version = value |
| 68 | + elif key == "NEEDRESTART-UCEXP": |
| 69 | + self.microcode_expected_version = value |
| 70 | + elif key == "NEEDRESTART-UCSTA": |
| 71 | + self.microcode_status = MicroCodeStatus(int(value)) |
| 72 | + # Count the others |
| 73 | + else: |
| 74 | + needrestart_counter.update({key}) |
| 75 | + |
| 76 | + self.services_count = needrestart_counter["NEEDRESTART-SVC"] |
| 77 | + self.containers_count = needrestart_counter["NEEDRESTART-CONT"] |
| 78 | + self.sessions_count = needrestart_counter["NEEDRESTART-SESS"] |
| 79 | + |
| 80 | + |
| 81 | +def write_timestamp(registry, needrestart_data): |
| 82 | + g = Gauge( |
| 83 | + "needrestart_timestamp_seconds", |
| 84 | + "information about the version and when it was last run", |
| 85 | + labelnames=["version"], |
| 86 | + registry=registry, |
| 87 | + ) |
| 88 | + g.labels(needrestart_data.version).set(needrestart_data.timestamp) |
| 89 | + |
| 90 | + |
| 91 | +def write_kernel(registry, needrestart_data): |
| 92 | + if needrestart_data.kernel_status: |
| 93 | + e = Gauge( |
| 94 | + "needrestart_kernel_status_info", |
| 95 | + "information about the kernel status", |
| 96 | + labelnames=["current", "expected"], |
| 97 | + registry=registry, |
| 98 | + ) |
| 99 | + e.labels( |
| 100 | + needrestart_data.kernel_current_version, |
| 101 | + needrestart_data.kernel_expected_version, |
| 102 | + ).set(needrestart_data.kernel_status.value) |
| 103 | + |
| 104 | + |
| 105 | +def write_microcode(registry, needrestart_data): |
| 106 | + if needrestart_data.microcode_status: |
| 107 | + e = Gauge( |
| 108 | + "needrestart_microcode_status_info", |
| 109 | + "information about the microcode status", |
| 110 | + labelnames=["current", "expected"], |
| 111 | + registry=registry, |
| 112 | + ) |
| 113 | + e.labels( |
| 114 | + needrestart_data.microcode_current_version, |
| 115 | + needrestart_data.microcode_expected_version, |
| 116 | + ).set(needrestart_data.microcode_status.value) |
| 117 | + |
| 118 | + |
| 119 | +def write_services(registry, needrestart_data): |
| 120 | + g = Gauge( |
| 121 | + "needrestart_services_total", |
| 122 | + "number of services requiring a restart", |
| 123 | + registry=registry, |
| 124 | + ) |
| 125 | + g.set(needrestart_data.services_count) |
| 126 | + |
| 127 | + |
| 128 | +def write_containers(registry, needrestart_data): |
| 129 | + g = Gauge( |
| 130 | + "needrestart_containers_total", |
| 131 | + "number of containers requiring a restart", |
| 132 | + registry=registry, |
| 133 | + ) |
| 134 | + g.set(needrestart_data.containers_count) |
| 135 | + |
| 136 | + |
| 137 | +def write_sessions(registry, needrestart_data): |
| 138 | + g = Gauge( |
| 139 | + "needrestart_sessions_total", |
| 140 | + "number of sessions requiring a restart", |
| 141 | + registry=registry, |
| 142 | + ) |
| 143 | + g.set(needrestart_data.sessions_count) |
| 144 | + |
| 145 | + |
| 146 | +def main(): |
| 147 | + registry = CollectorRegistry() |
| 148 | + |
| 149 | + try: |
| 150 | + needrestart_output = subprocess.run( |
| 151 | + ["needrestart", "-b"], capture_output=True, text=True, check=True |
| 152 | + ).stdout |
| 153 | + needrestart_data = NeedRestartData(needrestart_output) |
| 154 | + except subprocess.CalledProcessError as e: |
| 155 | + print(f"Error executing needrestart:\n{e}", file=sys.stderr) |
| 156 | + sys.exit(1) |
| 157 | + except Exception as e: |
| 158 | + print(f"An unexpected error occurred:\n{e}", file=sys.stderr) |
| 159 | + sys.exit(1) |
| 160 | + |
| 161 | + write_timestamp(registry, needrestart_data) |
| 162 | + write_kernel(registry, needrestart_data) |
| 163 | + write_microcode(registry, needrestart_data) |
| 164 | + write_services(registry, needrestart_data) |
| 165 | + write_containers(registry, needrestart_data) |
| 166 | + write_sessions(registry, needrestart_data) |
| 167 | + |
| 168 | + print(generate_latest(registry).decode(), end="") |
| 169 | + |
| 170 | + |
| 171 | +if __name__ == "__main__": |
| 172 | + main() |
0 commit comments