Update docs for os-healthcheck toolkit layout

2026-05-05 21:50:20 +00:00
parent 65c7c82f0f
commit c88428d092
8 changed files with 16 additions and 10 deletions
@@ -0,0 +1,29 @@
+#!/usr/bin/env bash
+set -o errexit
+set -o nounset
+set -o pipefail
+
+threshold="${1:-80}"
+
+if [[ ! "$threshold" =~ ^[0-9]+$ ]] || (( threshold < 1 || threshold > 100 )); then
+  printf 'CRITICAL: invalid threshold "%s"; provide an integer from 1 to 100\n' "$threshold" >&2
+  exit 2
+fi
+
+status=0
+warning_threshold=$(( threshold > 5 ? threshold - 5 : threshold ))
+
+while read -r filesystem size used avail use_percent mountpoint; do
+  usage="${use_percent%\%}"
+
+  if (( usage >= threshold )); then
+    printf 'CRITICAL: %s mounted on %s is %s used; threshold is %s%% (%s free)\n' "$filesystem" "$mountpoint" "$use_percent" "$threshold" "$avail"
+    status=1
+  elif (( usage >= warning_threshold )); then
+    printf 'WARNING: %s mounted on %s is %s used; threshold is %s%%\n' "$filesystem" "$mountpoint" "$use_percent" "$threshold"
+  else
+    printf 'OK: %s mounted on %s is %s used\n' "$filesystem" "$mountpoint" "$use_percent"
+  fi
+done < <(df -P -x tmpfs -x devtmpfs | awk 'NR > 1 {print $1, $2, $3, $4, $5, $6}')
+
+exit "$status"
@@ -0,0 +1,68 @@
+#!/usr/bin/env bash
+set -o errexit
+set -o nounset
+set -o pipefail
+
+section() {
+  printf '\n== %s ==\n' "$1"
+}
+
+run_or_warn() {
+  local description="$1"
+  shift
+
+  if command -v "$1" >/dev/null 2>&1; then
+    "$@" || printf 'WARNING: %s command failed\n' "$description"
+  else
+    printf 'WARNING: %s command not available\n' "$1"
+  fi
+}
+
+top_processes() {
+  local sort_key="$1"
+
+  if command -v ps >/dev/null 2>&1; then
+    ps -eo pid,ppid,comm,%cpu,%mem --sort="$sort_key" | head -n 11
+  else
+    printf 'WARNING: ps command not available\n'
+  fi
+}
+
+section "Host"
+hostname
+uptime
+
+section "OS"
+if [[ -r /etc/os-release ]]; then
+  . /etc/os-release
+  printf '%s\n' "${PRETTY_NAME:-Unknown Linux}"
+else
+  printf 'WARNING: /etc/os-release not readable\n'
+fi
+uname -r
+
+section "CPU Load"
+if [[ -r /proc/loadavg ]]; then
+  awk '{print "1m="$1, "5m="$2, "15m="$3}' /proc/loadavg
+else
+  uptime
+fi
+
+section "Memory"
+run_or_warn "memory usage" free -h
+
+section "Disk"
+run_or_warn "disk usage" df -h -x tmpfs -x devtmpfs
+
+section "Failed systemd Services"
+if command -v systemctl >/dev/null 2>&1; then
+  systemctl --failed --no-pager || true
+else
+  printf 'WARNING: systemctl command not available\n'
+fi
+
+section "Top CPU Processes"
+top_processes "-%cpu"
+
+section "Top Memory Processes"
+top_processes "-%mem"
@@ -0,0 +1,148 @@
+#!/usr/bin/env bash
+set -o errexit
+set -o nounset
+set -o pipefail
+
+target="${1:-}"
+status=0
+warnings=()
+criticals=()
+
+section() {
+  printf '\n[%s]\n' "$1"
+}
+
+warn() {
+  warnings+=("$1")
+  printf 'WARNING: %s\n' "$1"
+}
+
+critical() {
+  criticals+=("$1")
+  status=1
+  printf 'CRITICAL: %s\n' "$1"
+}
+
+have() {
+  command -v "$1" >/dev/null 2>&1
+}
+
+run_if_available() {
+  local command_name="$1"
+  shift
+
+  if have "$command_name"; then
+    "$@" || warn "$command_name command failed"
+  else
+    warn "$command_name command not available"
+  fi
+}
+
+section "LOCAL NETWORK"
+if have ip; then
+  ip addr || warn "ip addr command failed"
+  printf '\nRouting table:\n'
+  ip route || warn "ip route command failed"
+  printf '\nDefault gateway:\n'
+  if ! ip route show default; then
+    critical "default gateway not found"
+  elif ! ip route show default | grep -q '^default '; then
+    critical "default gateway not configured"
+  fi
+else
+  warn "ip command not available"
+fi
+
+section "INTERFACES"
+active_interfaces=0
+if have ip; then
+  ip -br link || warn "interface state query failed"
+  active_interfaces="$(ip -br link 2>/dev/null | awk '$2 == "UP" && $1 != "lo" {count++} END {print count+0}')"
+  if (( active_interfaces == 0 )); then
+    critical "no active non-loopback interface detected"
+  else
+    printf 'OK: %s active non-loopback interface(s) detected\n' "$active_interfaces"
+  fi
+else
+  warn "cannot inspect interface state without ip command"
+fi
+
+section "DNS"
+if [[ -r /etc/resolv.conf ]]; then
+  cat /etc/resolv.conf
+else
+  warn "/etc/resolv.conf not readable"
+fi
+
+dns_target="${target:-localhost}"
+if have getent; then
+  if getent hosts "$dns_target" >/dev/null 2>&1; then
+    printf 'OK: DNS resolution succeeded for %s\n' "$dns_target"
+    getent hosts "$dns_target"
+  else
+    critical "DNS resolution failed for ${dns_target}"
+  fi
+elif have nslookup; then
+  if nslookup "$dns_target"; then
+    printf 'OK: DNS resolution succeeded for %s\n' "$dns_target"
+  else
+    critical "DNS resolution failed for ${dns_target}"
+  fi
+else
+  warn "no DNS lookup tool available"
+fi
+
+section "CONNECTIVITY"
+if [[ -n "$target" ]]; then
+  if have ping; then
+    if ping -c 3 -W 2 "$target"; then
+      printf 'OK: ping succeeded for %s\n' "$target"
+    else
+      critical "ping failed for ${target}"
+    fi
+  else
+    warn "ping command not available"
+  fi
+
+  run_if_available traceroute traceroute "$target"
+
+  if have nc; then
+    if nc -vz -w 3 "$target" 443; then
+      printf 'OK: TCP 443 reachable on %s\n' "$target"
+    else
+      critical "TCP 443 connectivity failed for ${target}"
+    fi
+  elif have curl; then
+    if curl --head --silent --show-error --connect-timeout 5 "https://${target}" >/dev/null; then
+      printf 'OK: HTTPS connectivity succeeded for %s\n' "$target"
+    else
+      critical "HTTPS connectivity failed for ${target}"
+    fi
+  else
+    warn "no TCP connectivity test tool available (nc or curl)"
+  fi
+else
+  printf 'OK: no target provided; skipped remote connectivity checks\n'
+fi
+
+section "PORTS"
+if have ss; then
+  ss -tuln || warn "ss command failed"
+else
+  warn "ss command not available"
+fi
+
+section "SUMMARY"
+if (( ${#criticals[@]} > 0 )); then
+  printf 'CRITICAL: %s issue(s) detected\n' "${#criticals[@]}"
+fi
+
+if (( ${#warnings[@]} > 0 )); then
+  printf 'WARNING: %s warning(s) detected\n' "${#warnings[@]}"
+fi
+
+if (( status == 0 )); then
+  printf 'OK: no obvious DNS or connectivity problems detected\n'
+fi
+
+exit "$status"
@@ -0,0 +1,60 @@
+#!/usr/bin/env bash
+set -o errexit
+set -o nounset
+set -o pipefail
+
+services=("$@")
+
+service_exists() {
+  local service="$1"
+  systemctl list-unit-files "${service}.service" --no-legend 2>/dev/null | awk '{print $1}' | grep -qx "${service}.service"
+}
+
+pick_default_scheduler() {
+  if service_exists cron; then
+    printf 'cron'
+  elif service_exists crond; then
+    printf 'crond'
+  else
+    printf 'cron'
+  fi
+}
+
+pick_default_ssh() {
+  if service_exists sshd; then
+    printf 'sshd'
+  elif service_exists ssh; then
+    printf 'ssh'
+  else
+    printf 'sshd'
+  fi
+}
+
+if ! command -v systemctl >/dev/null 2>&1; then
+  printf 'CRITICAL: systemctl command not available; cannot check services\n' >&2
+  exit 1
+fi
+
+if (( ${#services[@]} == 0 )); then
+  services=("$(pick_default_ssh)" "$(pick_default_scheduler)")
+fi
+
+status=0
+
+for service in "${services[@]}"; do
+  if ! service_exists "$service"; then
+    printf 'CRITICAL: %s service not found\n' "$service"
+    status=1
+    continue
+  fi
+
+  if systemctl is-active --quiet "$service"; then
+    printf 'OK: %s is active\n' "$service"
+  else
+    state="$(systemctl is-active "$service" 2>/dev/null || true)"
+    printf 'CRITICAL: %s is %s\n' "$service" "${state:-unknown}"
+    status=1
+  fi
+done
+
+exit "$status"
@@ -0,0 +1,81 @@
+#!/usr/bin/env bash
+set -o errexit
+set -o nounset
+set -o pipefail
+
+host="$(hostname)"
+timestamp="$(date '+%Y-%m-%d_%H%M%S')"
+report="/tmp/system_report_${host}_${timestamp}.txt"
+
+section() {
+  printf '\n== %s ==\n' "$1"
+}
+
+run_or_warn() {
+  local description="$1"
+  shift
+
+  if command -v "$1" >/dev/null 2>&1; then
+    "$@" || printf 'WARNING: %s command failed\n' "$description"
+  else
+    printf 'WARNING: %s command not available\n' "$1"
+  fi
+}
+
+{
+  section "Host"
+  hostname
+
+  section "Date"
+  date
+
+  section "Uptime"
+  uptime
+
+  section "OS"
+  if [[ -r /etc/os-release ]]; then
+    . /etc/os-release
+    printf '%s\n' "${PRETTY_NAME:-Unknown Linux}"
+  else
+    printf 'WARNING: /etc/os-release not readable\n'
+  fi
+
+  section "Kernel"
+  uname -r
+
+  section "CPU Load"
+  if [[ -r /proc/loadavg ]]; then
+    awk '{print "1m="$1, "5m="$2, "15m="$3}' /proc/loadavg
+  else
+    uptime
+  fi
+
+  section "Memory"
+  run_or_warn "memory usage" free -h
+
+  section "Disk"
+  run_or_warn "disk usage" df -h -x tmpfs -x devtmpfs
+
+  section "Failed systemd Services"
+  if command -v systemctl >/dev/null 2>&1; then
+    systemctl --failed --no-pager || true
+  else
+    printf 'WARNING: systemctl command not available\n'
+  fi
+
+  section "Listening Ports"
+  if command -v ss >/dev/null 2>&1; then
+    ss -tuln || printf 'WARNING: ss command failed\n'
+  else
+    printf 'WARNING: ss command not available\n'
+  fi
+
+  section "Recent Kernel Messages"
+  if command -v journalctl >/dev/null 2>&1; then
+    journalctl -k -n 50 --no-pager || printf 'WARNING: journalctl kernel log query failed\n'
+  else
+    printf 'WARNING: journalctl command not available\n'
+  fi
+} > "$report"
+
+printf 'System report written to: %s\n' "$report"