#!/usr/bin/env bash
#
# Redis Sentinel Prometheus Metrics Exporter
#
# Prometheus textfile collector exporter for Redis Sentinel.
# Uses redis-cli to query Sentinel commands (SENTINEL masters, SENTINEL
# sentinels, SENTINEL slaves, INFO sentinel) and exports cluster health,
# master status, replica counts, quorum state, and failover statistics.
#
# Usage:
#   ./redis-sentinel-exporter.sh
#   ./redis-sentinel-exporter.sh --textfile
#   REDIS_HOST="sentinel.example.com" REDIS_PORT="26379" ./redis-sentinel-exporter.sh --textfile
#   ./redis-sentinel-exporter.sh --install
#
# Parameters:
#   --textfile   Write to textfile collector directory
#   --install    Create cron job for automatic collection
#   --help       Show usage
#
# Environment:
#   REDIS_HOST      Sentinel host (default: 127.0.0.1)
#   REDIS_PORT      Sentinel port (default: 26379)
#   REDIS_AUTH      Sentinel password (optional)
#   TEXTFILE_DIR    Textfile collector directory (default: /var/lib/node_exporter/textfile_collector)
#   CURL_TIMEOUT    Command timeout in seconds (default: 10)
#
# Author:  Phil Connor
# Contact: contact@mylinux.work
# Website: https://mylinux.work
# License: MIT
# Version: 1.0
#
# Metrics Exported:
#   Core:
#     - redis_sentinel_up
#     - redis_sentinel_exporter_info{version}
#     - redis_sentinel_masters_total
#
#   Per-Master:
#     - redis_sentinel_master_status{name}
#     - redis_sentinel_master_slaves{name}
#     - redis_sentinel_master_sentinels{name}
#     - redis_sentinel_master_quorum{name}
#     - redis_sentinel_master_odown{name}
#     - redis_sentinel_master_sdown{name}
#
#   Cluster:
#     - redis_sentinel_failover_total
#
#   Exporter:
#     - redis_sentinel_exporter_duration_seconds
#     - redis_sentinel_exporter_last_run_timestamp

set -euo pipefail

# --- Configuration ---
readonly VERSION="1.0"
readonly SCRIPT_NAME="$(basename "$0")"
REDIS_HOST="${REDIS_HOST:-127.0.0.1}"
REDIS_PORT="${REDIS_PORT:-26379}"
REDIS_AUTH="${REDIS_AUTH:-}"
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]

Redis Sentinel 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:
  REDIS_HOST      Sentinel host (default: 127.0.0.1)
  REDIS_PORT      Sentinel port (default: 26379)
  REDIS_AUTH      Sentinel password (optional)
  TEXTFILE_DIR    Output directory (default: /var/lib/node_exporter/textfile_collector)
  CURL_TIMEOUT    Command timeout in seconds (default: 10)

Examples:
  $SCRIPT_NAME
  $SCRIPT_NAME --textfile
  REDIS_HOST="sentinel.example.com" REDIS_PORT="26379" $SCRIPT_NAME --textfile
  REDIS_AUTH="secret" $SCRIPT_NAME --install
EOF
    exit 0
}

check_dependencies() {
    local missing=()
    for cmd in redis-cli awk; 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 "$REDIS_HOST" ]]; then
        echo "ERROR: REDIS_HOST is empty" >&2
        exit 1
    fi
    if [[ -z "$REDIS_PORT" ]]; then
        echo "ERROR: REDIS_PORT is empty" >&2
        exit 1
    fi
}

redis_cmd() {
    local args=(-h "$REDIS_HOST" -p "$REDIS_PORT" --no-auth-warning)
    if [[ -n "$REDIS_AUTH" ]]; then
        args+=(-a "$REDIS_AUTH")
    fi
    timeout "$CURL_TIMEOUT" redis-cli "${args[@]}" "$@" 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_sentinel() {
    # Test connectivity with PING
    local ping_result
    ping_result=$(redis_cmd PING)

    if [[ "$ping_result" != *"PONG"* ]]; then
        add_metric "redis_sentinel_up" "gauge" "Redis Sentinel reachability (1=up, 0=down)" "0"
        return 1
    fi

    add_metric "redis_sentinel_up" "gauge" "Redis Sentinel reachability (1=up, 0=down)" "1"

    # Get INFO sentinel for global stats
    local info_output
    info_output=$(redis_cmd INFO sentinel)

    local sentinel_masters=0
    if [[ -n "$info_output" ]]; then
        sentinel_masters=$(echo "$info_output" | awk -F: '/^sentinel_masters:/{gsub(/\r/,""); print $2}')
        sentinel_masters="${sentinel_masters:-0}"
    fi

    add_metric "redis_sentinel_masters_total" "gauge" "Total number of monitored masters" "$sentinel_masters"

    # Get SENTINEL masters for per-master details
    local masters_output
    masters_output=$(redis_cmd SENTINEL masters)

    if [[ -z "$masters_output" ]]; then
        return 0
    fi

    # Parse SENTINEL masters output (flat key-value pairs per master)
    local master_name="" flags="" num_slaves="" num_sentinels="" quorum=""
    local failover_total=0
    local first_status=true
    local first_slaves=true
    local first_sentinels=true
    local first_quorum=true
    local first_odown=true
    local first_sdown=true

    while IFS= read -r line; do
        line="${line%%$'\r'}"

        if [[ "$line" == "name" ]]; then
            # If we had a previous master, emit its metrics
            if [[ -n "$master_name" ]]; then
                _emit_master_metrics
            fi
            read -r master_name
            master_name="${master_name%%$'\r'}"
            flags=""
            num_slaves=""
            num_sentinels=""
            quorum=""
            continue
        fi

        case "$line" in
            flags) read -r flags; flags="${flags%%$'\r'}" ;;
            num-slaves) read -r num_slaves; num_slaves="${num_slaves%%$'\r'}" ;;
            num-other-sentinels) read -r num_sentinels; num_sentinels="${num_sentinels%%$'\r'}" ;;
            quorum) read -r quorum; quorum="${quorum%%$'\r'}" ;;
            failover-timeout) ;;
        esac
    done <<< "$masters_output"

    # Emit last master
    if [[ -n "$master_name" ]]; then
        _emit_master_metrics
    fi

    # Count total failovers from INFO sentinel
    local failover_count=0
    if [[ -n "$info_output" ]]; then
        # Sum running-scripts and scripts-queue-length as proxy, or parse sentinel_tilt/failover lines
        local running_scripts
        running_scripts=$(echo "$info_output" | awk -F: '/^sentinel_running_scripts:/{gsub(/\r/,""); print $2}')
        # Try to get failover count from master lines in INFO
        while IFS= read -r info_line; do
            info_line="${info_line%%$'\r'}"
            if [[ "$info_line" == master* ]]; then
                local fc
                fc=$(echo "$info_line" | grep -oP 'failovers=\K[0-9]+' || echo "0")
                failover_count=$((failover_count + fc))
            fi
        done <<< "$info_output"
    fi

    add_metric "redis_sentinel_failover_total" "gauge" "Total number of failovers observed" "$failover_count"

    return 0
}

_emit_master_metrics() {
    # Status: 1=ok, 0=down
    local status=1
    if [[ "$flags" == *"s_down"* ]] || [[ "$flags" == *"o_down"* ]]; then
        status=0
    fi

    if [[ "$first_status" == true ]]; then
        add_metric "redis_sentinel_master_status" "gauge" "Master status (1=ok, 0=down)" "$status" "name=\"${master_name}\""
        first_status=false
    else
        add_metric_value "redis_sentinel_master_status" "$status" "name=\"${master_name}\""
    fi

    # Slaves count
    if [[ "$first_slaves" == true ]]; then
        add_metric "redis_sentinel_master_slaves" "gauge" "Number of replicas for master" "${num_slaves:-0}" "name=\"${master_name}\""
        first_slaves=false
    else
        add_metric_value "redis_sentinel_master_slaves" "${num_slaves:-0}" "name=\"${master_name}\""
    fi

    # Sentinels count (add 1 for self)
    local total_sentinels=$(( ${num_sentinels:-0} + 1 ))
    if [[ "$first_sentinels" == true ]]; then
        add_metric "redis_sentinel_master_sentinels" "gauge" "Number of sentinels monitoring master" "$total_sentinels" "name=\"${master_name}\""
        first_sentinels=false
    else
        add_metric_value "redis_sentinel_master_sentinels" "$total_sentinels" "name=\"${master_name}\""
    fi

    # Quorum
    if [[ "$first_quorum" == true ]]; then
        add_metric "redis_sentinel_master_quorum" "gauge" "Quorum required for master" "${quorum:-0}" "name=\"${master_name}\""
        first_quorum=false
    else
        add_metric_value "redis_sentinel_master_quorum" "${quorum:-0}" "name=\"${master_name}\""
    fi

    # ODOWN
    local odown=0
    if [[ "$flags" == *"o_down"* ]]; then
        odown=1
    fi
    if [[ "$first_odown" == true ]]; then
        add_metric "redis_sentinel_master_odown" "gauge" "Master in objectively down state (1=yes, 0=no)" "$odown" "name=\"${master_name}\""
        first_odown=false
    else
        add_metric_value "redis_sentinel_master_odown" "$odown" "name=\"${master_name}\""
    fi

    # SDOWN
    local sdown=0
    if [[ "$flags" == *"s_down"* ]]; then
        sdown=1
    fi
    if [[ "$first_sdown" == true ]]; then
        add_metric "redis_sentinel_master_sdown" "gauge" "Master in subjectively down state (1=yes, 0=no)" "$sdown" "name=\"${master_name}\""
        first_sdown=false
    else
        add_metric_value "redis_sentinel_master_sdown" "$sdown" "name=\"${master_name}\""
    fi
}

write_output() {
    if [[ "$TEXTFILE_MODE" == true ]]; then
        local output_file="${TEXTFILE_DIR}/redis_sentinel.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/redis-sentinel-exporter <<EOF
# Redis Sentinel Prometheus Exporter — runs every 2 minutes
REDIS_HOST=${REDIS_HOST}
REDIS_PORT=${REDIS_PORT}
REDIS_AUTH=${REDIS_AUTH}
TEXTFILE_DIR=${TEXTFILE_DIR}
*/2 * * * * root ${script_path} --textfile 2>/dev/null
EOF

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

    # Collect metrics
    collect_sentinel

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

    write_output
}

main "$@"