#!/usr/bin/env bash
#
# Graylog Prometheus Metrics Exporter
#
# Prometheus textfile collector exporter for Graylog.
# Uses the Graylog REST API to collect node status, throughput,
# journal health, buffer usage, input/stream/index counts,
# and cluster leadership state.
#
# Usage:
#   GRAYLOG_URL="http://graylog.example.com:9000/api" GRAYLOG_TOKEN="abc123" ./graylog-exporter.sh
#   GRAYLOG_URL="http://graylog.example.com:9000/api" GRAYLOG_TOKEN="abc123" ./graylog-exporter.sh --textfile
#   GRAYLOG_URL="http://graylog.example.com:9000/api" GRAYLOG_TOKEN="abc123" ./graylog-exporter.sh --install
#
# Parameters:
#   --textfile   Write to textfile collector directory
#   --install    Create cron job for automatic collection
#   --help       Show usage
#
# Environment:
#   GRAYLOG_URL       Graylog API base URL (required, e.g. http://localhost:9000/api)
#   GRAYLOG_TOKEN     API token (required)
#   TEXTFILE_DIR      Textfile collector directory (default: /var/lib/node_exporter/textfile_collector)
#   CURL_TIMEOUT      API request timeout in seconds (default: 10)
#
# Author:  Phil Connor
# Contact: contact@mylinux.work
# Website: https://mylinux.work
# License: MIT
# Version: 1.0
#
# Metrics Exported:
#   Core:
#     - graylog_up
#     - graylog_exporter_info{version}
#     - graylog_node_is_leader
#     - graylog_input_count
#
#   Throughput:
#     - graylog_throughput_input
#     - graylog_throughput_output
#
#   Journal:
#     - graylog_journal_size_bytes
#     - graylog_journal_uncommitted_entries
#     - graylog_journal_events_read_count
#     - graylog_journal_events_append_count
#
#   Buffers:
#     - graylog_buffer_input_usage
#     - graylog_buffer_process_usage
#     - graylog_buffer_output_usage
#
#   Counts:
#     - graylog_stream_count
#     - graylog_index_count
#     - graylog_sidecar_count
#     - graylog_sidecar_active_count
#     - graylog_content_pack_count
#
#   Exporter:
#     - graylog_exporter_duration_seconds
#     - graylog_exporter_last_run_timestamp

set -euo pipefail

# --- Configuration ---
readonly VERSION="1.1"
readonly SCRIPT_NAME="$(basename "$0")"
GRAYLOG_URL="${GRAYLOG_URL:-}"
GRAYLOG_TOKEN="${GRAYLOG_TOKEN:-}"
TEXTFILE_DIR="${TEXTFILE_DIR:-/var/lib/node_exporter/textfile_collector}"
CURL_TIMEOUT="${CURL_TIMEOUT:-10}"
TEXTFILE_MODE=false
OUTPUT=""
START_TIME=""

# --- Functions ---

usage() {
    cat <<EOF
Usage: $SCRIPT_NAME [OPTIONS]

Graylog Prometheus Metrics Exporter

Options:
  --textfile    Write metrics to textfile collector directory
  --install     Create cron job for automatic collection
  --help        Show this help message

Environment Variables:
  GRAYLOG_URL       Graylog API base URL (required, e.g. http://localhost:9000/api)
  GRAYLOG_TOKEN     API token (required)
  TEXTFILE_DIR      Output directory (default: /var/lib/node_exporter/textfile_collector)
  CURL_TIMEOUT      Request timeout in seconds (default: 10)

Examples:
  GRAYLOG_URL="http://graylog.example.com:9000/api" GRAYLOG_TOKEN="abc123" $SCRIPT_NAME
  GRAYLOG_URL="http://graylog.example.com:9000/api" GRAYLOG_TOKEN="abc123" $SCRIPT_NAME --textfile
  GRAYLOG_URL="http://graylog.example.com:9000/api" GRAYLOG_TOKEN="abc123" $SCRIPT_NAME --install
EOF
    exit 0
}

check_dependencies() {
    local missing=()
    for cmd in curl jq; do
        if ! command -v "$cmd" &>/dev/null; then
            missing+=("$cmd")
        fi
    done
    if [[ ${#missing[@]} -gt 0 ]]; then
        echo "ERROR: Missing required commands: ${missing[*]}" >&2
        echo "Install with: apt install ${missing[*]}  OR  dnf install ${missing[*]}" >&2
        exit 1
    fi
}

validate_config() {
    if [[ -z "$GRAYLOG_URL" ]]; then
        echo "ERROR: GRAYLOG_URL environment variable is required" >&2
        exit 1
    fi
    if [[ -z "$GRAYLOG_TOKEN" ]]; then
        echo "ERROR: GRAYLOG_TOKEN environment variable is required" >&2
        exit 1
    fi
    # Strip trailing slash
    GRAYLOG_URL="${GRAYLOG_URL%/}"
}

api_get() {
    local endpoint="$1"
    curl -sf --max-time "$CURL_TIMEOUT" \
        -u "${GRAYLOG_TOKEN}:token" \
        -H "Accept: application/json" \
        "${GRAYLOG_URL}${endpoint}" 2>/dev/null || echo ""
}

add_metric() {
    local name="$1"
    local type="$2"
    local help="$3"
    local value="$4"
    local labels="${5:-}"

    if [[ -n "$labels" ]]; then
        OUTPUT+="# HELP ${name} ${help}
# TYPE ${name} ${type}
${name}{${labels}} ${value}
"
    else
        OUTPUT+="# HELP ${name} ${help}
# TYPE ${name} ${type}
${name} ${value}
"
    fi
}

add_metric_value() {
    local name="$1"
    local value="$2"
    local labels="${3:-}"

    if [[ -n "$labels" ]]; then
        OUTPUT+="${name}{${labels}} ${value}
"
    else
        OUTPUT+="${name} ${value}
"
    fi
}

collect_cluster() {
    local cluster_json
    cluster_json=$(api_get "/cluster")

    if [[ -z "$cluster_json" ]]; then
        add_metric "graylog_up" "gauge" "Graylog reachability (1=up, 0=down)" "0"
        return 1
    fi

    add_metric "graylog_up" "gauge" "Graylog reachability (1=up, 0=down)" "1"

    # Determine leader status from the first (local) node
    local is_leader
    is_leader=$(echo "$cluster_json" | jq -r '[.[]][0].is_leader // [.[]][0].is_master // false' 2>/dev/null)

    if [[ "$is_leader" == "true" ]]; then
        add_metric "graylog_node_is_leader" "gauge" "Whether this node is the cluster leader (1=leader, 0=follower)" "1"
    else
        add_metric "graylog_node_is_leader" "gauge" "Whether this node is the cluster leader (1=leader, 0=follower)" "0"
    fi

    return 0
}

collect_throughput() {
    local throughput_json
    throughput_json=$(api_get "/system/throughput")

    if [[ -z "$throughput_json" ]]; then
        return
    fi

    local input_throughput output_throughput
    input_throughput=$(echo "$throughput_json" | jq -r '.throughput.input // .throughput // 0' 2>/dev/null)
    output_throughput=$(echo "$throughput_json" | jq -r '.throughput.output // 0' 2>/dev/null)

    # Fallback: some Graylog versions expose a single throughput value
    if [[ "$input_throughput" == "0" || -z "$input_throughput" ]]; then
        input_throughput=$(echo "$throughput_json" | jq -r '.throughput // 0' 2>/dev/null)
    fi

    add_metric "graylog_throughput_input" "gauge" "Messages received per second" "${input_throughput:-0}"
    add_metric "graylog_throughput_output" "gauge" "Messages written per second" "${output_throughput:-0}"
}

collect_inputs() {
    local inputs_json
    inputs_json=$(api_get "/system/inputs")

    if [[ -z "$inputs_json" ]]; then
        return
    fi

    local input_count
    input_count=$(echo "$inputs_json" | jq -r '.total // 0' 2>/dev/null)
    add_metric "graylog_input_count" "gauge" "Total number of configured inputs" "${input_count:-0}"
}

collect_journal() {
    local journal_json
    journal_json=$(api_get "/system/journal")

    if [[ -z "$journal_json" ]]; then
        return
    fi

    local journal_size uncommitted read_events append_events
    journal_size=$(echo "$journal_json" | jq -r '.journal_size // 0' 2>/dev/null)
    uncommitted=$(echo "$journal_json" | jq -r '.uncommitted_journal_entries // 0' 2>/dev/null)
    read_events=$(echo "$journal_json" | jq -r '.read_events_per_second // 0' 2>/dev/null)
    append_events=$(echo "$journal_json" | jq -r '.append_events_per_second // 0' 2>/dev/null)

    add_metric "graylog_journal_size_bytes" "gauge" "Journal size in bytes" "${journal_size:-0}"
    add_metric "graylog_journal_uncommitted_entries" "gauge" "Uncommitted journal entries" "${uncommitted:-0}"
    add_metric "graylog_journal_events_read_count" "gauge" "Journal events read per second" "${read_events:-0}"
    add_metric "graylog_journal_events_append_count" "gauge" "Journal events appended per second" "${append_events:-0}"
}

collect_buffers() {
    local buffers_json
    buffers_json=$(api_get "/system/buffers")

    if [[ -z "$buffers_json" ]]; then
        return
    fi

    local input_usage process_usage output_usage
    input_usage=$(echo "$buffers_json" | jq -r '.buffers.input.utilization_percent // 0' 2>/dev/null)
    process_usage=$(echo "$buffers_json" | jq -r '.buffers.process.utilization_percent // 0' 2>/dev/null)
    output_usage=$(echo "$buffers_json" | jq -r '.buffers.output.utilization_percent // 0' 2>/dev/null)

    add_metric "graylog_buffer_input_usage" "gauge" "Input buffer utilization percentage" "${input_usage:-0}"
    add_metric "graylog_buffer_process_usage" "gauge" "Process buffer utilization percentage" "${process_usage:-0}"
    add_metric "graylog_buffer_output_usage" "gauge" "Output buffer utilization percentage" "${output_usage:-0}"
}

collect_streams() {
    local streams_json
    streams_json=$(api_get "/streams")

    if [[ -z "$streams_json" ]]; then
        return
    fi

    local stream_count
    stream_count=$(echo "$streams_json" | jq -r '.total // 0' 2>/dev/null)
    add_metric "graylog_stream_count" "gauge" "Total number of streams" "${stream_count:-0}"
}

collect_indices() {
    local indices_json
    indices_json=$(api_get "/system/indexer/indices")

    if [[ -z "$indices_json" ]]; then
        return
    fi

    local index_count
    index_count=$(echo "$indices_json" | jq -r '.all.indices | length // 0' 2>/dev/null)

    # Fallback: try counting top-level keys
    if [[ "$index_count" == "0" || -z "$index_count" ]]; then
        index_count=$(echo "$indices_json" | jq -r 'if type == "object" then keys | length else 0 end' 2>/dev/null)
    fi

    add_metric "graylog_index_count" "gauge" "Total number of indices" "${index_count:-0}"
}

collect_sidecars() {
    local sidecars_json
    sidecars_json=$(api_get "/sidecars")

    if [[ -z "$sidecars_json" ]]; then
        return
    fi

    local total_count active_count
    total_count=$(echo "$sidecars_json" | jq -r '.pagination.total // 0' 2>/dev/null)
    active_count=$(echo "$sidecars_json" | jq -r '[.sidecars[] | select(.node_details.status.status == 1)] | length // 0' 2>/dev/null)

    add_metric "graylog_sidecar_count" "gauge" "Total number of registered Sidecars" "${total_count:-0}"
    add_metric "graylog_sidecar_active_count" "gauge" "Number of active Sidecars" "${active_count:-0}"
}

collect_content_packs() {
    local packs_json
    packs_json=$(api_get "/system/content_packs/latest")

    if [[ -z "$packs_json" ]]; then
        return
    fi

    local pack_count
    pack_count=$(echo "$packs_json" | jq -r '.total // 0' 2>/dev/null)
    add_metric "graylog_content_pack_count" "gauge" "Total number of content packs" "${pack_count:-0}"
}

write_output() {
    if [[ "$TEXTFILE_MODE" == true ]]; then
        local output_file="${TEXTFILE_DIR}/graylog.prom"
        local temp_file="${output_file}.$$"

        mkdir -p "$TEXTFILE_DIR"
        echo "$OUTPUT" > "$temp_file"
        mv "$temp_file" "$output_file"
    else
        echo "$OUTPUT"
    fi
}

install_cron() {
    if [[ $EUID -ne 0 ]]; then
        echo "ERROR: --install requires root" >&2
        exit 1
    fi

    local script_path
    script_path=$(readlink -f "$0")

    cat > /etc/cron.d/graylog-exporter <<EOF
# Graylog Prometheus Exporter — runs every 2 minutes
GRAYLOG_URL=${GRAYLOG_URL}
GRAYLOG_TOKEN=${GRAYLOG_TOKEN}
TEXTFILE_DIR=${TEXTFILE_DIR}
*/2 * * * * root ${script_path} --textfile 2>/dev/null
EOF

    chmod 644 /etc/cron.d/graylog-exporter
    echo "Installed cron job: /etc/cron.d/graylog-exporter"
    echo "Metrics will be written to: ${TEXTFILE_DIR}/graylog.prom"
}

# --- Main ---

main() {
    # Parse arguments
    for arg in "$@"; do
        case "$arg" in
            --textfile) TEXTFILE_MODE=true ;;
            --install)
                check_dependencies
                validate_config
                install_cron
                exit 0
                ;;
            --help|-h) usage ;;
            *) echo "Unknown option: $arg" >&2; usage ;;
        esac
    done

    check_dependencies
    validate_config

    START_TIME=$(date +%s%N)

    # Exporter info
    add_metric "graylog_exporter_info" "gauge" "Exporter version information" "1" "version=\"${VERSION}\""

    # Collect metrics
    if collect_cluster; then
        collect_inputs
        collect_throughput
        collect_journal
        collect_buffers
        collect_streams
        collect_indices
        collect_sidecars
        collect_content_packs
    fi

    # Exporter performance
    local end_time duration
    end_time=$(date +%s%N)
    duration=$(echo "scale=2; ($end_time - $START_TIME) / 1000000000" | bc 2>/dev/null || echo "0")
    add_metric "graylog_exporter_duration_seconds" "gauge" "Time to generate all metrics" "$duration"
    add_metric "graylog_exporter_last_run_timestamp" "gauge" "Unix timestamp of last successful run" "$(date +%s)"

    write_output
}

main "$@"