portfolio/infra-run/scripts/bash/os-healthcheck/healthcheck.sh

#!/usr/bin/env bash
set -euo 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"