#!/bin/bash
################################################################################
# Script Name: textfile-health-exporter.sh
# Version: 1.0
# Description: Prometheus exporter that monitors node_exporter textfile
#              collector .prom files — detects stale files, parse errors,
#              file sizes, and missing cron jobs. A meta-exporter that
#              watches the health of your other exporters.
#
# Author: Phil Connor
# Contact: contact@mylinux.work
# Website: https://mylinux.work
# License: MIT
#
# Prerequisites:
#   - awk, stat, find
#   - netcat (nc) for HTTP mode
#
# Usage:
#   ./textfile-health-exporter.sh                    # stdout
#   ./textfile-health-exporter.sh --textfile          # node_exporter textfile
#   ./textfile-health-exporter.sh --http              # HTTP server on port 9202
#
# Metrics Exported:
#   Per-File:
#     - textfile_health_file_age_seconds{file}        - Seconds since last modified
#     - textfile_health_file_size_bytes{file}          - File size in bytes
#     - textfile_health_file_lines{file}               - Line count
#     - textfile_health_parse_ok{file}                 - 1 if valid, 0 if errors
#
#   Summary:
#     - textfile_health_files_total                    - Total .prom files found
#     - textfile_health_stale_files_total              - Files older than threshold
#     - textfile_health_parse_errors_total             - Files with parse errors
#     - textfile_health_total_size_bytes               - Total size of all files
#
#   Exporter:
#     - textfile_health_up                             - Exporter status (1=up)
#     - textfile_health_stale_threshold_seconds        - Configured stale threshold
#     - textfile_health_duration_seconds               - Script execution time
#     - textfile_health_last_run_timestamp             - Last run timestamp
#
# Configuration:
#   Default HTTP port: 9202
#   Textfile directory: /var/lib/node_exporter
#   Stale threshold: 600 seconds (10 minutes)
#
################################################################################

set -uo pipefail

# ============================================================================
# CONFIGURATION
# ============================================================================

TEXTFILE_DIR="/var/lib/node_exporter"
OUTPUT_FILE=""
HTTP_MODE=false
HTTP_PORT=9202
STALE_THRESHOLD=600
OWN_OUTPUT_FILE="textfile_health.prom"

# ============================================================================
# USAGE
# ============================================================================

show_usage() {
    cat <<EOF
Usage: $(basename "$0") [OPTIONS]

Monitor node_exporter textfile collector .prom files for staleness and errors.

MODES:
    --textfile      Write to node_exporter textfile collector
    --http          Run HTTP server on port ${HTTP_PORT}
    (default)       Print metrics to stdout

OPTIONS:
    -p, --port N              HTTP port (default: ${HTTP_PORT})
    -o, --output F            Output file path
    -d, --dir PATH            Textfile directory (default: ${TEXTFILE_DIR})
    -s, --stale-threshold N   Stale threshold in seconds (default: ${STALE_THRESHOLD})
    -h, --help                Show this help

EXAMPLES:
    $(basename "$0") --textfile
    $(basename "$0") --http --port 9202
    $(basename "$0") --dir /var/lib/node_exporter --stale-threshold 300
EOF
    exit 0
}

# ============================================================================
# ARGUMENT PARSING
# ============================================================================

parse_args() {
    while [[ $# -gt 0 ]]; do
        case "$1" in
            -h|--help)              show_usage ;;
            --textfile)             OUTPUT_FILE="${TEXTFILE_DIR}/${OWN_OUTPUT_FILE}"; shift ;;
            --http)                 HTTP_MODE=true; shift ;;
            -p|--port)              HTTP_PORT="$2"; shift 2 ;;
            -o|--output)            OUTPUT_FILE="$2"; shift 2 ;;
            -d|--dir)               TEXTFILE_DIR="$2"; shift 2 ;;
            -s|--stale-threshold)   STALE_THRESHOLD="$2"; shift 2 ;;
            *)                      echo "# ERROR: Unknown option: $1" >&2; show_usage ;;
        esac
    done
}

# ============================================================================
# PREFLIGHT
# ============================================================================

preflight() {
    local missing=()
    for cmd in awk stat find wc; do
        command -v "$cmd" &>/dev/null || missing+=("$cmd")
    done
    if [[ ${#missing[@]} -gt 0 ]]; then
        echo "# ERROR: Missing required commands: ${missing[*]}" >&2
        exit 1
    fi
    if [[ ! -d "$TEXTFILE_DIR" ]]; then
        echo "# ERROR: Textfile directory does not exist: ${TEXTFILE_DIR}" >&2
        exit 1
    fi
}

# ============================================================================
# PARSE VALIDATION
# ============================================================================

validate_prom_file() {
    local file="$1"
    awk '
    /^[[:space:]]*$/ { next }
    /^#/ { next }
    /^[a-zA-Z_:][a-zA-Z0-9_:]*(\{[^}]*\})?[[:space:]]+-?[0-9]/ { next }
    /^[a-zA-Z_:][a-zA-Z0-9_:]*(\{[^}]*\})?[[:space:]]+[+-]?[Ii][Nn][Ff]/ { next }
    /^[a-zA-Z_:][a-zA-Z0-9_:]*(\{[^}]*\})?[[:space:]]+[Nn][Aa][Nn]/ { next }
    { errors++; }
    END { exit (errors > 0) ? 1 : 0 }
    ' "$file" 2>/dev/null
}

# ============================================================================
# METRICS COLLECTION
# ============================================================================

collect_metrics() {
    local start_time
    start_time=$(date +%s%N)
    local now
    now=$(date +%s)

    echo "# HELP textfile_health_up Exporter status (1=up)"
    echo "# TYPE textfile_health_up gauge"
    echo "textfile_health_up 1"
    echo ""

    echo "# HELP textfile_health_stale_threshold_seconds Configured stale threshold"
    echo "# TYPE textfile_health_stale_threshold_seconds gauge"
    echo "textfile_health_stale_threshold_seconds ${STALE_THRESHOLD}"
    echo ""

    # --- Resolve own output filename for exclusion ---
    local exclude_basename
    if [[ -n "$OUTPUT_FILE" ]]; then
        exclude_basename=$(basename "$OUTPUT_FILE")
    else
        exclude_basename="$OWN_OUTPUT_FILE"
    fi

    # --- Find all .prom files, excluding own output ---
    local prom_files=()
    while IFS= read -r -d '' f; do
        local base
        base=$(basename "$f")
        [[ "$base" == "$exclude_basename" ]] && continue
        prom_files+=("$f")
    done < <(find "$TEXTFILE_DIR" -maxdepth 1 -name '*.prom' -type f -print0 2>/dev/null)

    local files_total=${#prom_files[@]}
    local stale_total=0
    local parse_errors_total=0
    local total_size=0

    # --- Per-file metrics ---
    if [[ $files_total -gt 0 ]]; then
        echo "# HELP textfile_health_file_age_seconds Seconds since file was last modified"
        echo "# TYPE textfile_health_file_age_seconds gauge"
        for f in "${prom_files[@]}"; do
            local fname
            fname=$(basename "$f")
            local mtime
            mtime=$(stat -c %Y "$f" 2>/dev/null) || continue
            local age=$(( now - mtime ))
            echo "textfile_health_file_age_seconds{file=\"${fname}\"} ${age}"
        done
        echo ""

        echo "# HELP textfile_health_file_size_bytes File size in bytes"
        echo "# TYPE textfile_health_file_size_bytes gauge"
        for f in "${prom_files[@]}"; do
            local fname
            fname=$(basename "$f")
            local size
            size=$(stat -c %s "$f" 2>/dev/null) || continue
            echo "textfile_health_file_size_bytes{file=\"${fname}\"} ${size}"
            total_size=$(( total_size + size ))
        done
        echo ""

        echo "# HELP textfile_health_file_lines Line count per file"
        echo "# TYPE textfile_health_file_lines gauge"
        for f in "${prom_files[@]}"; do
            local fname
            fname=$(basename "$f")
            local lines
            lines=$(wc -l < "$f" 2>/dev/null) || continue
            echo "textfile_health_file_lines{file=\"${fname}\"} ${lines}"
        done
        echo ""

        echo "# HELP textfile_health_parse_ok Parse validation (1=valid, 0=errors)"
        echo "# TYPE textfile_health_parse_ok gauge"
        for f in "${prom_files[@]}"; do
            local fname
            fname=$(basename "$f")
            if validate_prom_file "$f"; then
                echo "textfile_health_parse_ok{file=\"${fname}\"} 1"
            else
                echo "textfile_health_parse_ok{file=\"${fname}\"} 0"
                parse_errors_total=$(( parse_errors_total + 1 ))
            fi
        done
        echo ""

        # --- Count stale files ---
        for f in "${prom_files[@]}"; do
            local mtime
            mtime=$(stat -c %Y "$f" 2>/dev/null) || continue
            local age=$(( now - mtime ))
            if [[ $age -ge $STALE_THRESHOLD ]]; then
                stale_total=$(( stale_total + 1 ))
            fi
        done
    fi

    # --- Summary metrics ---
    echo "# HELP textfile_health_files_total Total .prom files found"
    echo "# TYPE textfile_health_files_total gauge"
    echo "textfile_health_files_total ${files_total}"
    echo ""

    echo "# HELP textfile_health_stale_files_total Files older than stale threshold"
    echo "# TYPE textfile_health_stale_files_total gauge"
    echo "textfile_health_stale_files_total ${stale_total}"
    echo ""

    echo "# HELP textfile_health_parse_errors_total Files with parse errors"
    echo "# TYPE textfile_health_parse_errors_total gauge"
    echo "textfile_health_parse_errors_total ${parse_errors_total}"
    echo ""

    echo "# HELP textfile_health_total_size_bytes Total size of all .prom files"
    echo "# TYPE textfile_health_total_size_bytes gauge"
    echo "textfile_health_total_size_bytes ${total_size}"
    echo ""

    # --- Exporter metadata ---
    local end_time duration
    end_time=$(date +%s%N)
    duration=$(echo "scale=3; ($end_time - $start_time) / 1000000000" | bc 2>/dev/null || echo "0")

    echo "# HELP textfile_health_duration_seconds Script execution time"
    echo "# TYPE textfile_health_duration_seconds gauge"
    echo "textfile_health_duration_seconds ${duration}"
    echo ""

    echo "# HELP textfile_health_last_run_timestamp Last successful run (unix timestamp)"
    echo "# TYPE textfile_health_last_run_timestamp gauge"
    echo "textfile_health_last_run_timestamp ${now}"
}

# ============================================================================
# OUTPUT HANDLING
# ============================================================================

output_metrics() {
    local metrics
    metrics=$(collect_metrics)

    if [[ -n "$OUTPUT_FILE" ]]; then
        local output_dir
        output_dir="$(dirname "$OUTPUT_FILE")"
        mkdir -p "$output_dir"

        local temp_file
        temp_file=$(mktemp "${output_dir}/.textfile_health.XXXXXX")

        echo "$metrics" > "$temp_file"

        local file_lines
        file_lines=$(wc -l < "$temp_file" 2>/dev/null || echo 0)

        if [[ "$file_lines" -lt 5 ]]; then
            rm -f "$temp_file"
            echo "# ERROR: Metrics file too small (${file_lines} lines), keeping previous" >&2
            exit 1
        fi

        chmod 644 "$temp_file"
        mv -f "$temp_file" "$OUTPUT_FILE"
        echo "# Metrics written to ${OUTPUT_FILE} (${file_lines} lines)" >&2
    else
        echo "$metrics"
    fi
}

serve_http() {
    echo "# Starting HTTP server on port ${HTTP_PORT}" >&2
    echo "# Metrics endpoint: http://localhost:${HTTP_PORT}/metrics" >&2

    if ! command -v nc &>/dev/null && ! command -v ncat &>/dev/null; then
        echo "# ERROR: netcat (nc/ncat) required for HTTP mode" >&2
        exit 1
    fi

    local nc_cmd="nc"
    command -v ncat &>/dev/null && nc_cmd="ncat"

    while true; do
        local metrics
        metrics=$(collect_metrics)
        local content_length=${#metrics}
        local response="HTTP/1.1 200 OK\r\nContent-Type: text/plain; charset=utf-8\r\nContent-Length: ${content_length}\r\nConnection: close\r\n\r\n${metrics}"
        echo -e "$response" | $nc_cmd -l -p "$HTTP_PORT" -q 1 2>/dev/null || \
        echo -e "$response" | $nc_cmd -l "$HTTP_PORT" 2>/dev/null || true
    done
}

# ============================================================================
# MAIN
# ============================================================================

parse_args "$@"
preflight

if [[ "$HTTP_MODE" == "true" ]]; then
    serve_http
else
    output_metrics
fi