#!/usr/bin/env bash
#
# Jenkins Prometheus Metrics Exporter
#
# Prometheus textfile collector exporter for Jenkins.
# Uses the Jenkins JSON API to collect build queue length, executor
# utilization, job success/failure rates, build duration, node status,
# disk usage, and JVM heap statistics.
#
# Usage:
#   JENKINS_URL="https://jenkins.example.com" JENKINS_USER="admin" JENKINS_TOKEN="xxx" ./jenkins-exporter.sh
#   JENKINS_URL="https://jenkins.example.com" JENKINS_USER="admin" JENKINS_TOKEN="xxx" ./jenkins-exporter.sh --textfile
#   JENKINS_URL="https://jenkins.example.com" JENKINS_USER="admin" JENKINS_TOKEN="xxx" ./jenkins-exporter.sh --install
#
# Parameters:
#   --textfile   Write to textfile collector directory
#   --install    Create cron job for automatic collection
#   --help       Show usage
#
# Environment:
#   JENKINS_URL     Jenkins base URL (required)
#   JENKINS_USER    Jenkins username (required)
#   JENKINS_TOKEN   Jenkins 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:
#     - jenkins_up
#     - jenkins_exporter_info{version}
#
#   Build Queue:
#     - jenkins_build_queue_length
#
#   Executors:
#     - jenkins_executors_total
#     - jenkins_executors_busy
#     - jenkins_executors_idle
#
#   Jobs:
#     - jenkins_jobs_total
#     - jenkins_job_last_build_duration_seconds{job}
#     - jenkins_job_last_build_result{job}
#
#   Nodes:
#     - jenkins_nodes_total
#     - jenkins_nodes_online
#     - jenkins_nodes_offline
#
#   Disk:
#     - jenkins_disk_usage_bytes
#
#   JVM:
#     - jenkins_jvm_heap_used_bytes
#     - jenkins_jvm_heap_max_bytes
#
#   Exporter:
#     - jenkins_exporter_duration_seconds
#     - jenkins_exporter_last_run_timestamp

set -euo pipefail

# --- Configuration ---
readonly VERSION="1.0"
readonly SCRIPT_NAME="$(basename "$0")"
JENKINS_URL="${JENKINS_URL:-}"
JENKINS_USER="${JENKINS_USER:-}"
JENKINS_TOKEN="${JENKINS_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]

Jenkins 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:
  JENKINS_URL     Jenkins base URL (required)
  JENKINS_USER    Jenkins username (required)
  JENKINS_TOKEN   Jenkins API token (required)
  TEXTFILE_DIR    Output directory (default: /var/lib/node_exporter/textfile_collector)
  CURL_TIMEOUT    Request timeout in seconds (default: 10)

Examples:
  JENKINS_URL="https://jenkins.example.com" JENKINS_USER="admin" JENKINS_TOKEN="xxx" $SCRIPT_NAME
  JENKINS_URL="https://jenkins.example.com" JENKINS_USER="admin" JENKINS_TOKEN="xxx" $SCRIPT_NAME --textfile
  JENKINS_URL="https://jenkins.example.com" JENKINS_USER="admin" JENKINS_TOKEN="xxx" $SCRIPT_NAME --install
EOF
    exit 0
}

check_dependencies() {
    local missing=()
    for cmd in curl jq bc; 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 "$JENKINS_URL" ]]; then
        echo "ERROR: JENKINS_URL environment variable is required" >&2
        exit 1
    fi
    if [[ -z "$JENKINS_USER" ]]; then
        echo "ERROR: JENKINS_USER environment variable is required" >&2
        exit 1
    fi
    if [[ -z "$JENKINS_TOKEN" ]]; then
        echo "ERROR: JENKINS_TOKEN environment variable is required" >&2
        exit 1
    fi
    # Strip trailing slash
    JENKINS_URL="${JENKINS_URL%/}"
}

api_get() {
    local endpoint="$1"
    curl -sf --max-time "$CURL_TIMEOUT" \
        -u "${JENKINS_USER}:${JENKINS_TOKEN}" \
        "${JENKINS_URL}${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_health() {
    local api_json
    api_json=$(api_get "/api/json")

    if [[ -z "$api_json" ]]; then
        add_metric "jenkins_up" "gauge" "Jenkins reachability (1=up, 0=down)" "0"
        return 1
    fi

    add_metric "jenkins_up" "gauge" "Jenkins reachability (1=up, 0=down)" "1"
    return 0
}

collect_queue() {
    local queue_json
    queue_json=$(api_get "/queue/api/json")

    if [[ -n "$queue_json" ]]; then
        local queue_length
        queue_length=$(echo "$queue_json" | jq -r '.items | length // 0' 2>/dev/null)
        add_metric "jenkins_build_queue_length" "gauge" "Number of items in the build queue" "${queue_length:-0}"
    fi
}

collect_executors() {
    local computer_json
    computer_json=$(api_get "/computer/api/json")

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

    local total_executors busy_executors idle_executors
    total_executors=$(echo "$computer_json" | jq -r '.totalExecutors // 0' 2>/dev/null)
    busy_executors=$(echo "$computer_json" | jq -r '.busyExecutors // 0' 2>/dev/null)
    idle_executors=$((total_executors - busy_executors))

    add_metric "jenkins_executors_total" "gauge" "Total number of executors" "${total_executors:-0}"
    add_metric "jenkins_executors_busy" "gauge" "Number of busy executors" "${busy_executors:-0}"
    add_metric "jenkins_executors_idle" "gauge" "Number of idle executors" "${idle_executors:-0}"

    # Node status
    local nodes_total nodes_online nodes_offline
    nodes_total=$(echo "$computer_json" | jq -r '.computer | length // 0' 2>/dev/null)
    nodes_offline=$(echo "$computer_json" | jq -r '[.computer[] | select(.offline == true)] | length // 0' 2>/dev/null)
    nodes_online=$((nodes_total - nodes_offline))

    add_metric "jenkins_nodes_total" "gauge" "Total number of nodes" "${nodes_total:-0}"
    add_metric "jenkins_nodes_online" "gauge" "Number of online nodes" "${nodes_online:-0}"
    add_metric "jenkins_nodes_offline" "gauge" "Number of offline nodes" "${nodes_offline:-0}"

    # JVM stats from the master node monitor data
    local heap_used heap_max
    heap_used=$(echo "$computer_json" | jq -r '.computer[0].monitorData["hudson.node_monitors.SwapSpaceMonitor"]["totalPhysicalMemory"] // empty' 2>/dev/null)

    # Use the built-in master node's memory monitor
    local master_monitor
    master_monitor=$(echo "$computer_json" | jq -r '.computer[] | select(.displayName == "Built-In Node" or .displayName == "master") | .monitorData // empty' 2>/dev/null)

    if [[ -n "$master_monitor" ]]; then
        heap_used=$(echo "$master_monitor" | jq -r '.["hudson.node_monitors.SwapSpaceMonitor"]["availablePhysicalMemory"] // empty' 2>/dev/null)
        heap_max=$(echo "$master_monitor" | jq -r '.["hudson.node_monitors.SwapSpaceMonitor"]["totalPhysicalMemory"] // empty' 2>/dev/null)
    fi

    # Try the overallLoad API for JVM info
    local overall_json
    overall_json=$(api_get "/overallLoad/api/json")

    if [[ -n "$overall_json" ]]; then
        # overallLoad provides queue length stats as well
        :
    fi

    # Collect JVM heap from the system properties or groovy script endpoint
    # Fall back to the computer API memory data
    if [[ -n "${heap_used:-}" && "$heap_used" != "null" ]]; then
        add_metric "jenkins_jvm_heap_used_bytes" "gauge" "JVM heap memory used" "$heap_used"
    fi
    if [[ -n "${heap_max:-}" && "$heap_max" != "null" ]]; then
        add_metric "jenkins_jvm_heap_max_bytes" "gauge" "JVM heap memory max" "$heap_max"
    fi
}

collect_jobs() {
    local jobs_json
    jobs_json=$(api_get "/api/json?tree=jobs[name,lastBuild[duration,result]]")

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

    local job_count
    job_count=$(echo "$jobs_json" | jq -r '.jobs | length // 0' 2>/dev/null)
    add_metric "jenkins_jobs_total" "gauge" "Total number of jobs" "${job_count:-0}"

    # Per-job metrics
    local job_names
    job_names=$(echo "$jobs_json" | jq -r '.jobs[]? | .name // empty' 2>/dev/null)

    local first_duration=true
    local first_result=true

    while IFS= read -r job_name; do
        [[ -z "$job_name" ]] && continue

        local duration result
        duration=$(echo "$jobs_json" | jq -r --arg name "$job_name" '.jobs[] | select(.name == $name) | .lastBuild.duration // empty' 2>/dev/null)
        result=$(echo "$jobs_json" | jq -r --arg name "$job_name" '.jobs[] | select(.name == $name) | .lastBuild.result // empty' 2>/dev/null)

        # Sanitise job name for Prometheus labels
        local safe_name
        safe_name=$(echo "$job_name" | sed 's/[^a-zA-Z0-9_\-]/_/g')

        if [[ -n "$duration" && "$duration" != "null" ]]; then
            local duration_seconds
            duration_seconds=$(echo "scale=2; $duration / 1000" | bc 2>/dev/null || echo "0")
            if [[ "$first_duration" == true ]]; then
                add_metric "jenkins_job_last_build_duration_seconds" "gauge" "Last build duration per job" "$duration_seconds" "job=\"${safe_name}\""
                first_duration=false
            else
                add_metric_value "jenkins_job_last_build_duration_seconds" "$duration_seconds" "job=\"${safe_name}\""
            fi
        fi

        if [[ -n "$result" && "$result" != "null" ]]; then
            local result_value
            if [[ "$result" == "SUCCESS" ]]; then
                result_value=1
            else
                result_value=0
            fi
            if [[ "$first_result" == true ]]; then
                add_metric "jenkins_job_last_build_result" "gauge" "Last build result (1=SUCCESS, 0=FAILURE)" "$result_value" "job=\"${safe_name}\""
                first_result=false
            else
                add_metric_value "jenkins_job_last_build_result" "$result_value" "job=\"${safe_name}\""
            fi
        fi
    done <<< "$job_names"
}

collect_disk_usage() {
    local disk_json
    disk_json=$(api_get "/disk-usage/api/json")

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

    local disk_usage
    disk_usage=$(echo "$disk_json" | jq -r '.diskUsage // empty' 2>/dev/null)

    if [[ -n "$disk_usage" && "$disk_usage" != "null" ]]; then
        add_metric "jenkins_disk_usage_bytes" "gauge" "Total disk usage (disk-usage plugin)" "$disk_usage"
    fi
}

write_output() {
    if [[ "$TEXTFILE_MODE" == true ]]; then
        local output_file="${TEXTFILE_DIR}/jenkins.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/jenkins-exporter <<EOF
# Jenkins Prometheus Exporter — runs every 2 minutes
JENKINS_URL=${JENKINS_URL}
JENKINS_USER=${JENKINS_USER}
JENKINS_TOKEN=${JENKINS_TOKEN}
TEXTFILE_DIR=${TEXTFILE_DIR}
*/2 * * * * root ${script_path} --textfile 2>/dev/null
EOF

    chmod 644 /etc/cron.d/jenkins-exporter
    echo "Installed cron job: /etc/cron.d/jenkins-exporter"
    echo "Metrics will be written to: ${TEXTFILE_DIR}/jenkins.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 "jenkins_exporter_info" "gauge" "Exporter version information" "1" "version=\"${VERSION}\""

    # Collect metrics
    if collect_health; then
        collect_queue
        collect_executors
        collect_jobs
        collect_disk_usage
    fi

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

    write_output
}

main "$@"