#!/usr/bin/env bash
#
# Consul Prometheus Metrics Exporter
#
# Prometheus textfile collector exporter for Consul.
# Uses the Consul HTTP API to collect cluster health, Raft consensus,
# service catalog, health check states, KV store entry counts,
# and node membership.
#
# Usage:
#   ./consul-exporter.sh
#   ./consul-exporter.sh --textfile
#   CONSUL_TOKEN="xxx" ./consul-exporter.sh --textfile
#   ./consul-exporter.sh --install
#
# Parameters:
#   --textfile   Write to textfile collector directory
#   --install    Create cron job for automatic collection
#   --help       Show usage
#
# Environment:
#   CONSUL_URL      Consul HTTP API base URL (default: http://127.0.0.1:8500)
#   CONSUL_TOKEN    ACL token (optional, required if ACLs are enabled)
#   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:
#     - consul_up
#     - consul_exporter_info{version}
#     - consul_peers_total
#     - consul_leader
#
#   Catalog:
#     - consul_services_total
#     - consul_nodes_total
#
#   Health:
#     - consul_health_checks_passing
#     - consul_health_checks_warning
#     - consul_health_checks_critical
#
#   KV:
#     - consul_kv_entries_total
#
#   Raft:
#     - consul_raft_commit_time_seconds
#     - consul_raft_last_contact_seconds
#
#   Exporter:
#     - consul_exporter_duration_seconds
#     - consul_exporter_last_run_timestamp

set -euo pipefail

# --- Configuration ---
readonly VERSION="1.0"
readonly SCRIPT_NAME="$(basename "$0")"
CONSUL_URL="${CONSUL_URL:-http://127.0.0.1:8500}"
CONSUL_TOKEN="${CONSUL_TOKEN:-}"
TEXTFILE_DIR="${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]

Consul 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:
  CONSUL_URL      Consul HTTP API base URL (default: http://127.0.0.1:8500)
  CONSUL_TOKEN    ACL token (optional)
  TEXTFILE_DIR    Output directory (default: /var/lib/node_exporter/textfile_collector)
  CURL_TIMEOUT    Request timeout in seconds (default: 10)

Examples:
  $SCRIPT_NAME
  CONSUL_TOKEN="xxx" $SCRIPT_NAME --textfile
  CONSUL_URL="http://consul.example.com:8500" $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() {
    # Strip trailing slash
    CONSUL_URL="${CONSUL_URL%/}"
}

api_get() {
    local endpoint="$1"
    local curl_args=(-sf --max-time "$CURL_TIMEOUT")

    if [[ -n "$CONSUL_TOKEN" ]]; then
        curl_args+=(-H "X-Consul-Token: ${CONSUL_TOKEN}")
    fi

    curl "${curl_args[@]}" "${CONSUL_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 members_json
    members_json=$(api_get "/v1/agent/members")

    if [[ -z "$members_json" ]]; then
        add_metric "consul_up" "gauge" "Consul reachability (1=up, 0=down)" "0"
        return 1
    fi

    add_metric "consul_up" "gauge" "Consul reachability (1=up, 0=down)" "1"

    # Node count from members
    local node_count
    node_count=$(echo "$members_json" | jq 'length' 2>/dev/null)
    add_metric "consul_nodes_total" "gauge" "Total number of cluster nodes" "${node_count:-0}"

    return 0
}

collect_raft() {
    local raft_json
    raft_json=$(api_get "/v1/operator/raft/configuration")

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

    # Peer count
    local peer_count
    peer_count=$(echo "$raft_json" | jq '.Servers | length' 2>/dev/null)
    add_metric "consul_peers_total" "gauge" "Number of Raft peers in the cluster" "${peer_count:-0}"

    # Leader detection — check if current node is leader
    local self_json leader_addr self_addr
    self_json=$(api_get "/v1/agent/self")

    if [[ -n "$self_json" ]]; then
        leader_addr=$(echo "$raft_json" | jq -r '.Servers[] | select(.Leader == true) | .Address' 2>/dev/null)
        self_addr=$(echo "$self_json" | jq -r '.Config.RaftAddress // .Stats.raft.applied_index // empty' 2>/dev/null)
        local self_name self_leader_name
        self_name=$(echo "$self_json" | jq -r '.Config.NodeName // empty' 2>/dev/null)
        self_leader_name=$(echo "$raft_json" | jq -r '.Servers[] | select(.Leader == true) | .Node' 2>/dev/null)

        if [[ -n "$self_name" && "$self_name" == "$self_leader_name" ]]; then
            add_metric "consul_leader" "gauge" "Whether this node is the cluster leader (1=leader, 0=follower)" "1"
        else
            add_metric "consul_leader" "gauge" "Whether this node is the cluster leader (1=leader, 0=follower)" "0"
        fi

        # Raft stats from /v1/agent/self
        local raft_commit_time raft_last_contact
        raft_commit_time=$(echo "$self_json" | jq -r '.Stats.raft.commit_time // empty' 2>/dev/null)
        raft_last_contact=$(echo "$self_json" | jq -r '.Stats.raft.last_contact // empty' 2>/dev/null)

        if [[ -n "$raft_commit_time" ]]; then
            # Convert milliseconds to seconds
            local commit_seconds
            commit_seconds=$(echo "scale=6; ${raft_commit_time%ms} / 1000" | bc 2>/dev/null || echo "0")
            add_metric "consul_raft_commit_time_seconds" "gauge" "Raft commit time in seconds" "$commit_seconds"
        fi

        if [[ -n "$raft_last_contact" ]]; then
            # Convert milliseconds to seconds
            local contact_seconds
            contact_seconds=$(echo "scale=6; ${raft_last_contact%ms} / 1000" | bc 2>/dev/null || echo "0")
            add_metric "consul_raft_last_contact_seconds" "gauge" "Time since last Raft leader contact in seconds" "$contact_seconds"
        fi
    fi
}

collect_services() {
    local services_json
    services_json=$(api_get "/v1/catalog/services")

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

    local service_count
    service_count=$(echo "$services_json" | jq 'keys | length' 2>/dev/null)
    add_metric "consul_services_total" "gauge" "Total number of registered services" "${service_count:-0}"
}

collect_health_checks() {
    local checks_json
    checks_json=$(api_get "/v1/health/state/any")

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

    local passing warning critical
    passing=$(echo "$checks_json" | jq '[.[] | select(.Status == "passing")] | length' 2>/dev/null)
    warning=$(echo "$checks_json" | jq '[.[] | select(.Status == "warning")] | length' 2>/dev/null)
    critical=$(echo "$checks_json" | jq '[.[] | select(.Status == "critical")] | length' 2>/dev/null)

    add_metric "consul_health_checks_passing" "gauge" "Number of passing health checks" "${passing:-0}"
    add_metric "consul_health_checks_warning" "gauge" "Number of warning health checks" "${warning:-0}"
    add_metric "consul_health_checks_critical" "gauge" "Number of critical health checks" "${critical:-0}"
}

collect_kv() {
    local kv_json
    kv_json=$(api_get "/v1/kv/?keys")

    if [[ -z "$kv_json" ]]; then
        add_metric "consul_kv_entries_total" "gauge" "Total number of KV store entries" "0"
        return
    fi

    local kv_count
    kv_count=$(echo "$kv_json" | jq 'length' 2>/dev/null)
    add_metric "consul_kv_entries_total" "gauge" "Total number of KV store entries" "${kv_count:-0}"
}

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

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

    # Collect metrics
    if collect_health; then
        collect_raft
        collect_services
        collect_health_checks
        collect_kv
    fi

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

    write_output
}

main "$@"