#!/bin/bash

################################################
#### Redis Metrics Collector                ####
#### for Prometheus node_exporter textfile  ####
####                                        ####
#### Author: Phil Connor                    ####
#### Contact: contact@mylinux.work          ####
#### Version: 1.0.0.20260309               ####
################################################

set -o pipefail

SCRIPT_NAME=$(basename "$0")
readonly SCRIPT_NAME

# Default configuration
readonly DEFAULT_NODE_DIR="/var/lib/node_exporter"
readonly DEFAULT_COLLECTION_INTERVAL=60
readonly DEFAULT_TOP_N_KEYS=10

# Configuration variables (can be overridden by environment)
REDIS_HOST=${REDIS_HOST:-127.0.0.1}
REDIS_PORT=${REDIS_PORT:-6379}
REDIS_PASSWORD=${REDIS_PASSWORD:-}
REDIS_CLI=${REDIS_CLI:-}
NODE_DIR=${NODE_DIR:-$DEFAULT_NODE_DIR}
COLLECTION_INTERVAL=${COLLECTION_INTERVAL:-$DEFAULT_COLLECTION_INTERVAL}
TOP_N_KEYS=${TOP_N_KEYS:-$DEFAULT_TOP_N_KEYS}
DEBUG=${DEBUG:-}

# Runtime flags
RUN_MODE="once"

# Error tracking
ERRORS_TOTAL=0

handle_error() {
    local exit_code=$1
    local line_number=$2
    echo "Error: $SCRIPT_NAME failed at line $line_number with exit code $exit_code" >&2
    exit "$exit_code"
}

trap 'handle_error $? $LINENO' ERR

debug_echo() {
    if [[ -n "$DEBUG" ]]; then
        echo "[DEBUG] $*" >&2
    fi
}

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

Redis metrics collector for Prometheus node_exporter textfile directory.

Collects connected clients, memory usage, keyspace hit/miss ratios, per-database
key counts, slowlog analysis, big key detection, persistence status, replication
info, and uptime from Redis via redis-cli.

OPTIONS:
    --once          Run collection once and exit (default)
    --daemon        Run continuously at COLLECTION_INTERVAL
    --help, -h      Show this help message

ENVIRONMENT VARIABLES:
    REDIS_HOST              Redis server hostname (default: 127.0.0.1)
    REDIS_PORT              Redis server port (default: 6379)
    REDIS_PASSWORD          Redis AUTH password (default: empty)
    REDIS_CLI               Path to redis-cli binary (default: auto-detect)
    NODE_DIR                Node exporter textfile directory (default: $DEFAULT_NODE_DIR)
    COLLECTION_INTERVAL     Seconds between collections in daemon mode (default: $DEFAULT_COLLECTION_INTERVAL)
    TOP_N_KEYS              Number of largest keys to report (default: $DEFAULT_TOP_N_KEYS)
    DEBUG                   Enable debug output

EXAMPLES:
    $SCRIPT_NAME --once
    REDIS_PASSWORD=secret $SCRIPT_NAME --daemon
    REDIS_HOST=redis.example.com REDIS_PORT=6380 $SCRIPT_NAME

OUTPUT:
    Writes metrics to \$NODE_DIR/textfile_collector/redis_metrics.prom

EOF
    exit 0
}

# Parse arguments
while [[ $# -gt 0 ]]; do
    case "$1" in
        --once)     RUN_MODE="once"; shift ;;
        --daemon)   RUN_MODE="daemon"; shift ;;
        --help|-h)  show_help ;;
        *) echo "Unknown option: $1" >&2; show_help ;;
    esac
done

# Auto-detect redis-cli
detect_redis_cli() {
    if [[ -n "$REDIS_CLI" ]]; then
        if [[ ! -x "$REDIS_CLI" ]]; then
            echo "Error: REDIS_CLI not found or not executable: $REDIS_CLI" >&2
            exit 1
        fi
        return
    fi

    REDIS_CLI=$(command -v redis-cli 2>/dev/null) || true
    if [[ -z "$REDIS_CLI" ]]; then
        echo "Error: redis-cli not found in PATH" >&2
        exit 1
    fi
}

# Validate configuration
validate_config() {
    detect_redis_cli

    local textfile_dir="${NODE_DIR}/textfile_collector"
    if [[ ! -d "$textfile_dir" ]]; then
        echo "Error: Textfile collector directory not found: $textfile_dir" >&2
        echo "Create it: sudo mkdir -p $textfile_dir" >&2
        exit 1
    fi
}

# Run a redis-cli command
redis_cmd() {
    local cmd="$*"
    local args=(-h "$REDIS_HOST" -p "$REDIS_PORT")

    if [[ -n "$REDIS_PASSWORD" ]]; then
        args+=(-a "$REDIS_PASSWORD" --no-auth-warning)
    fi

    debug_echo "redis-cli ${args[*]} $cmd"

    $REDIS_CLI "${args[@]}" $cmd 2>/dev/null
}

# Collect metrics from INFO command
collect_info_metrics() {
    debug_echo "Collecting INFO metrics..."

    local info
    info=$(redis_cmd INFO) || { ERRORS_TOTAL=$((ERRORS_TOTAL + 1)); return 1; }

    local connected_clients used_memory keyspace_hits keyspace_misses
    local rdb_last_save_time aof_enabled connected_slaves uptime_in_seconds

    connected_clients=$(echo "$info" | grep -oP '^connected_clients:\K[0-9]+' || echo "0")
    used_memory=$(echo "$info" | grep -oP '^used_memory:\K[0-9]+' || echo "0")
    keyspace_hits=$(echo "$info" | grep -oP '^keyspace_hits:\K[0-9]+' || echo "0")
    keyspace_misses=$(echo "$info" | grep -oP '^keyspace_misses:\K[0-9]+' || echo "0")
    rdb_last_save_time=$(echo "$info" | grep -oP '^rdb_last_save_time:\K[0-9]+' || echo "0")
    aof_enabled=$(echo "$info" | grep -oP '^aof_enabled:\K[0-9]+' || echo "0")
    connected_slaves=$(echo "$info" | grep -oP '^connected_slaves:\K[0-9]+' || echo "0")
    uptime_in_seconds=$(echo "$info" | grep -oP '^uptime_in_seconds:\K[0-9]+' || echo "0")

    cat <<EOF
# HELP redis_connected_clients Number of client connections.
# TYPE redis_connected_clients gauge
redis_connected_clients $connected_clients

# HELP redis_used_memory_bytes Total number of bytes allocated by Redis.
# TYPE redis_used_memory_bytes gauge
redis_used_memory_bytes $used_memory

# HELP redis_keyspace_hits_total Number of successful lookups of keys in the main dictionary.
# TYPE redis_keyspace_hits_total counter
redis_keyspace_hits_total $keyspace_hits

# HELP redis_keyspace_misses_total Number of failed lookups of keys in the main dictionary.
# TYPE redis_keyspace_misses_total counter
redis_keyspace_misses_total $keyspace_misses

# HELP redis_last_rdb_save_timestamp Unix timestamp of last successful RDB save.
# TYPE redis_last_rdb_save_timestamp gauge
redis_last_rdb_save_timestamp $rdb_last_save_time

# HELP redis_aof_enabled Whether AOF persistence is enabled (1=yes, 0=no).
# TYPE redis_aof_enabled gauge
redis_aof_enabled $aof_enabled

# HELP redis_connected_slaves Number of connected replicas.
# TYPE redis_connected_slaves gauge
redis_connected_slaves $connected_slaves

# HELP redis_uptime_seconds Number of seconds since Redis server start.
# TYPE redis_uptime_seconds gauge
redis_uptime_seconds $uptime_in_seconds

EOF
}

# Collect per-database key counts
collect_dbsize_metrics() {
    debug_echo "Collecting per-database key counts..."

    echo "# HELP redis_db_keys Number of keys in each Redis database."
    echo "# TYPE redis_db_keys gauge"

    for db in $(seq 0 15); do
        local dbsize
        dbsize=$(redis_cmd -n "$db" DBSIZE 2>/dev/null | grep -oP '[0-9]+' || echo "0")
        echo "redis_db_keys{db=\"$db\"} $dbsize"
    done

    echo ""
}

# Collect slowlog metrics
collect_slowlog_metrics() {
    debug_echo "Collecting slowlog metrics..."

    local slowlog_len
    slowlog_len=$(redis_cmd SLOWLOG LEN 2>/dev/null | grep -oP '[0-9]+' || echo "0")

    local slowlog_latest_duration=0
    if [[ "$slowlog_len" -gt 0 ]]; then
        local slowlog_entry
        slowlog_entry=$(redis_cmd SLOWLOG GET 1 2>/dev/null)
        if [[ -n "$slowlog_entry" ]]; then
            slowlog_latest_duration=$(echo "$slowlog_entry" | awk 'NR==4 { gsub(/[^0-9]/, ""); print }' || echo "0")
            [[ -z "$slowlog_latest_duration" ]] && slowlog_latest_duration=0
        fi
    fi

    cat <<EOF
# HELP redis_slowlog_length Total number of entries in the slowlog.
# TYPE redis_slowlog_length gauge
redis_slowlog_length $slowlog_len

# HELP redis_slowlog_latest_duration_microseconds Duration of the latest slowlog entry in microseconds.
# TYPE redis_slowlog_latest_duration_microseconds gauge
redis_slowlog_latest_duration_microseconds $slowlog_latest_duration

EOF
}

# Collect big key metrics via RANDOMKEY sampling
collect_big_keys() {
    debug_echo "Collecting big key metrics (sampling $TOP_N_KEYS keys)..."

    local -A key_sizes
    local sample_count=$((TOP_N_KEYS * 10))

    for ((i = 0; i < sample_count; i++)); do
        local key
        key=$(redis_cmd RANDOMKEY 2>/dev/null)
        [[ -z "$key" || "$key" == "(nil)" ]] && continue

        local size
        size=$(redis_cmd MEMORY USAGE "$key" 2>/dev/null | grep -oP '[0-9]+' || echo "0")
        [[ "$size" -gt 0 ]] && key_sizes["$key"]=$size
    done

    if [[ ${#key_sizes[@]} -gt 0 ]]; then
        echo "# HELP redis_big_key_bytes Memory usage in bytes of sampled large keys."
        echo "# TYPE redis_big_key_bytes gauge"

        for key in "${!key_sizes[@]}"; do
            echo "$key ${key_sizes[$key]}"
        done | sort -t' ' -k2 -rn | head -n "$TOP_N_KEYS" | while IFS=' ' read -r k v; do
            local safe_key
            safe_key=$(echo "$k" | sed 's/["\\]/\\&/g')
            echo "redis_big_key_bytes{key=\"$safe_key\"} $v"
        done

        echo ""
    fi
}

# Write collection metadata
collect_metadata() {
    cat <<EOF
# HELP redis_exporter_errors_total Total number of errors during collection.
# TYPE redis_exporter_errors_total counter
redis_exporter_errors_total $ERRORS_TOTAL

# HELP redis_exporter_last_success_timestamp Unix timestamp of last successful collection.
# TYPE redis_exporter_last_success_timestamp gauge
redis_exporter_last_success_timestamp $(date +%s)

EOF
}

# Main collection function
collect_all() {
    ERRORS_TOTAL=0
    local output_dir="${NODE_DIR}/textfile_collector"
    local output_file="${output_dir}/redis_metrics.prom"
    local temp_file
    temp_file=$(mktemp "${output_file}.XXXXXX")

    debug_echo "Starting collection..."

    {
        collect_info_metrics
        collect_dbsize_metrics
        collect_slowlog_metrics
        collect_big_keys
        collect_metadata
    } > "$temp_file" 2>/dev/null

    mv "$temp_file" "$output_file"

    debug_echo "Collection complete. Wrote to $output_file (errors: $ERRORS_TOTAL)"
}

# Main
main() {
    validate_config

    case "$RUN_MODE" in
        once)
            collect_all
            ;;
        daemon)
            echo "$SCRIPT_NAME running in daemon mode (interval: ${COLLECTION_INTERVAL}s)"
            while true; do
                collect_all
                sleep "$COLLECTION_INTERVAL"
            done
            ;;
    esac
}

main