mb-admin/scripts/install_redis.sh

#!/bin/bash
#
# install_redis.sh
# Install and configure Redis server for WordPress on LLSMP/AlmaLinux.
#
# Usage:
#   bash install_redis.sh              -- full install+configure flow
#   bash install_redis.sh --status     -- status check only (no install)
#
# Flow:
#   1. Check if Redis is already installed/running -> report and exit early if healthy
#   2. Install redis package via dnf (idempotent)
#   3. Locate and patch redis.conf for Unix socket
#   4. Grant litespeed user group access to the socket
#   5. Enable + start the service, wait for socket readiness
#   6. Verify PING via socket
#   7. Configure WordPress LiteSpeed Cache Plugin Object Cache
#

set -uo pipefail
# NOTE: Not using set -e globally so we can handle each error explicitly
# and produce meaningful messages instead of silent exits.

# ── Colors & helpers ─────────────────────────────────────────────────────────
RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[1;33m'
BLUE='\033[0;34m'
NC='\033[0m'

info()      { printf "${BLUE}[INFO] %s${NC}\n" "$*"; }
success()   { printf "${GREEN}[SUCCESS] %s${NC}\n" "$*"; }
warning()   { printf "${YELLOW}[WARNING] %s${NC}\n" "$*"; }
error_exit(){ printf "${RED}[ERROR] %s${NC}\n" "$*" >&2; exit 1; }

# ── Constants ────────────────────────────────────────────────────────────────
REDIS_SOCKET="/var/run/redis/redis.sock"
REDIS_SOCKET_DIR="/var/run/redis"
REDIS_CONF_PATHS=("/etc/redis/redis.conf" "/etc/redis.conf" "/etc/redis6/redis.conf" "/etc/redis7/redis.conf")
WP_ROOT="/var/www/webroot/ROOT"
WEB_USER="litespeed"
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
STATUS_ONLY=false

# ── Parse arguments ──────────────────────────────────────────────────────────
for arg in "$@"; do
    case "$arg" in
        --status) STATUS_ONLY=true ;;
    esac
done

# ── Helper: detect redis service name ────────────────────────────────────────
get_redis_svc() {
    systemctl list-unit-files --type=service 2>/dev/null \
        | grep -oP '[\w@-]*redis[\w@-]*\.service' \
        | grep -v 'sentinel' \
        | head -1 \
        | sed 's/\.service//'
}

# ── Helper: detect redis.conf path ───────────────────────────────────────────
get_redis_conf() {
    for path in "${REDIS_CONF_PATHS[@]}"; do
        if [[ -f "$path" ]]; then
            echo "$path"
            return 0
        fi
    done
    # Last resort: ask rpm
    local rpm_conf
    rpm_conf=$(rpm -ql redis 2>/dev/null | grep '\.conf$' | head -1)
    if [[ -n "$rpm_conf" && -f "$rpm_conf" ]]; then
        echo "$rpm_conf"
        return 0
    fi
    echo ""
}

# ── Helper: test redis socket connectivity ───────────────────────────────────
test_redis_socket() {
    if [[ ! -S "$REDIS_SOCKET" ]]; then
        return 1
    fi
    local result
    result=$(redis-cli -s "$REDIS_SOCKET" ping 2>/dev/null || true)
    [[ "$result" == "PONG" ]]
}

# ── STATUS CHECK ─────────────────────────────────────────────────────────────
do_status() {
    local svc
    svc=$(get_redis_svc)

    if [[ -z "$svc" ]]; then
        echo "Redis: NOT INSTALLED"
        return 1  # signal caller that install is needed
    fi

    if systemctl is-active --quiet "$svc"; then
        if test_redis_socket; then
            success "Redis: RUNNING (service=$svc, socket=$REDIS_SOCKET, ping=PONG)"
        else
            warning "Redis: RUNNING (service=$svc) but socket unreachable at $REDIS_SOCKET"
        fi
        return 0
    else
        local state
        state=$(systemctl is-enabled "$svc" 2>/dev/null || echo "unknown")
        warning "Redis: INSTALLED but STOPPED (service=$svc, enabled=$state)"
        return 2  # signal caller: installed but stopped
    fi
}

# ── STEP 1: Install redis package ────────────────────────────────────────────
step_install() {
    if command -v redis-server >/dev/null 2>&1; then
        info "Redis binary already present: $(command -v redis-server)"
        return 0
    fi

    info "Redis not found. Installing via dnf..."

    # Guard: check if dnf lock is held (another process installing)
    local lock_retries=5
    while fuser /var/lib/rpm/.rpm.lock >/dev/null 2>&1; do
        if [[ $lock_retries -le 0 ]]; then
            error_exit "dnf/rpm lock held by another process after 25 seconds. Try again later."
        fi
        warning "dnf/rpm lock detected, waiting 5s... ($lock_retries retries left)"
        sleep 5
        lock_retries=$((lock_retries - 1))
    done

    if ! dnf install -y redis 2>&1; then
        error_exit "dnf install redis failed. Check: (1) internet connectivity, (2) dnf repo config, (3) run: dnf install -y redis"
    fi

    # Verify binary appeared post-install
    if ! command -v redis-server >/dev/null 2>&1; then
        error_exit "dnf reported success but redis-server binary not found. Package may be broken."
    fi

    success "Redis package installed successfully."
}

# ── STEP 2: Configure redis.conf ─────────────────────────────────────────────
step_configure() {
    local conf
    conf=$(get_redis_conf)

    if [[ -z "$conf" ]]; then
        error_exit "Cannot find redis.conf after installation. Checked: ${REDIS_CONF_PATHS[*]}. Run: rpm -ql redis | grep conf"
    fi

    info "Using Redis config: $conf"

    # Backup (idempotent — only on first run)
    if [[ ! -f "${conf}.orig" ]]; then
        cp "$conf" "${conf}.orig" || error_exit "Cannot backup $conf — check disk space and permissions."
        info "Backed up original config to ${conf}.orig"
    fi

    # Ensure socket directory exists with correct ownership BEFORE writing config
    if ! mkdir -p "$REDIS_SOCKET_DIR" 2>/dev/null; then
        error_exit "Cannot create socket directory $REDIS_SOCKET_DIR"
    fi
    # redis user must own the socket dir so it can create the socket file
    if ! chown redis:redis "$REDIS_SOCKET_DIR" 2>/dev/null; then
        warning "Could not chown $REDIS_SOCKET_DIR to redis:redis — socket creation may fail."
    fi
    chmod 755 "$REDIS_SOCKET_DIR"

    # bind: restrict to loopback only (security hardening)
    if grep -q '^bind ' "$conf"; then
        sed -i 's/^bind .*/bind 127.0.0.1/' "$conf"
    else
        echo "bind 127.0.0.1" >> "$conf"
    fi

    # unixsocket path
    if grep -q '^unixsocket ' "$conf"; then
        sed -i "s|^unixsocket .*|unixsocket $REDIS_SOCKET|" "$conf"
    else
        echo "unixsocket $REDIS_SOCKET" >> "$conf"
    fi

    # unixsocketperm: 775 so redis group members (litespeed) can r/w
    if grep -q '^unixsocketperm ' "$conf"; then
        sed -i 's/^unixsocketperm .*/unixsocketperm 775/' "$conf"
    else
        echo "unixsocketperm 775" >> "$conf"
    fi

    # maxmemory: only set if not already defined
    if ! grep -q '^maxmemory ' "$conf"; then
        echo "maxmemory 256mb" >> "$conf"
        echo "maxmemory-policy allkeys-lru" >> "$conf"
        info "Set maxmemory 256mb with allkeys-lru eviction policy."
    fi

    # Verify config was written (sanity check)
    if ! grep -q "^unixsocket $REDIS_SOCKET" "$conf"; then
        error_exit "Config write verification failed — unixsocket not found in $conf after editing."
    fi

    success "Redis configuration applied at $conf"
}

# ── STEP 3: Add litespeed to redis group ─────────────────────────────────────
step_socket_access() {
    if ! id "$WEB_USER" >/dev/null 2>&1; then
        warning "User '$WEB_USER' does not exist on this system — skipping group membership."
        return 0
    fi

    if ! getent group redis >/dev/null 2>&1; then
        warning "Group 'redis' does not exist yet — will retry after service start."
        return 0
    fi

    if id -nG "$WEB_USER" | grep -qw redis; then
        info "$WEB_USER is already a member of the redis group."
        return 0
    fi

    if usermod -aG redis "$WEB_USER"; then
        success "$WEB_USER added to redis group."
    else
        warning "usermod failed to add $WEB_USER to redis group. Socket access may be restricted."
    fi
}

# ── STEP 4: Enable + start Redis service ─────────────────────────────────────
step_start() {
    local svc
    svc=$(get_redis_svc)

    if [[ -z "$svc" ]]; then
        error_exit "No Redis systemd service found after installation. Check: systemctl list-unit-files | grep redis"
    fi

    info "Enabling $svc at boot..."
    systemctl enable "$svc" 2>/dev/null || warning "Could not enable $svc at boot (non-fatal)."

    info "Starting $svc..."
    if ! systemctl restart "$svc" 2>/dev/null; then
        local journal
        journal=$(journalctl -u "$svc" -n 20 --no-pager 2>/dev/null || echo "unavailable")
        error_exit "Failed to start $svc. Last 20 journal lines:\n$journal"
    fi

    # Wait up to 15s for socket to appear
    local waited=0
    while [[ $waited -lt 15 ]]; do
        if [[ -S "$REDIS_SOCKET" ]]; then
            break
        fi
        sleep 1
        waited=$((waited + 1))
    done

    if [[ ! -S "$REDIS_SOCKET" ]]; then
        local journal
        journal=$(journalctl -u "$svc" -n 20 --no-pager 2>/dev/null || echo "unavailable")
        error_exit "Redis started ($svc) but socket not found at $REDIS_SOCKET after 15s.\nJournal:\n$journal"
    fi

    # Final PING verification
    local ping_result
    ping_result=$(redis-cli -s "$REDIS_SOCKET" ping 2>/dev/null || echo "FAILED")
    if [[ "$ping_result" != "PONG" ]]; then
        error_exit "Socket exists at $REDIS_SOCKET but PING returned '$ping_result'. Redis may have crashed."
    fi

    success "Redis is running and healthy (service=$svc, socket=$REDIS_SOCKET, ping=PONG)"
}

# ── STEP 5: Re-check group membership after service starts (socket now exists) ──
step_verify_access() {
    if ! id "$WEB_USER" >/dev/null 2>&1; then
        return 0
    fi

    # Ensure redis group exists now (created by service start)
    if ! getent group redis >/dev/null 2>&1; then
        warning "redis group still not found after service start — skipping access check."
        return 0
    fi

    if ! id -nG "$WEB_USER" | grep -qw redis; then
        info "Re-attempting to add $WEB_USER to redis group..."
        usermod -aG redis "$WEB_USER" || warning "Could not add $WEB_USER to redis group."
    fi

    # Verify socket is accessible by litespeed group
    local sock_perms
    sock_perms=$(stat -c '%A %G' "$REDIS_SOCKET" 2>/dev/null || echo "unknown")
    info "Socket permissions: $sock_perms"
}

# ── STEP 6: Configure WordPress Object Cache ──────────────────────────────────
step_wp_object_cache() {
    local plugin_script="$SCRIPT_DIR/configure_litespeed_plugin_object_cache.sh"

    # Guard: script exists
    if [[ ! -f "$plugin_script" ]]; then
        warning "configure_litespeed_plugin_object_cache.sh not found at $plugin_script — skipping WP config."
        return 0
    fi

    # Guard: WordPress present
    if [[ ! -f "$WP_ROOT/wp-config.php" ]]; then
        warning "WordPress not found at $WP_ROOT — skipping WP object cache config."
        return 0
    fi

    # Guard: WP-CLI available — resolve full path so sudo -u retains it
    local wp_bin
    for _wp in "/usr/local/bin/wp" "/usr/bin/wp" "/home/litespeed/bin/wp"; do
        if [ -x "$_wp" ]; then wp_bin="$_wp"; break; fi
    done
    [ -z "${wp_bin:-}" ] && wp_bin=$(command -v wp 2>/dev/null || echo "")
    if [ -z "${wp_bin:-}" ]; then
        warning "WP-CLI not found — skipping WP object cache config. Expected at /usr/local/bin/wp"
        return 0
    fi

    # Guard: LiteSpeed Cache plugin active (run as litespeed with full wp path)
    local plugin_status
    plugin_status=$(sudo -u "$WEB_USER" "$wp_bin" plugin is-active litespeed-cache --path="$WP_ROOT" 2>/dev/null && echo "active" || echo "inactive")
    if [[ "$plugin_status" != "active" ]]; then
        warning "LiteSpeed Cache plugin is not active in WordPress — skipping object cache config."
        warning "Activate it at: WordPress Admin > Plugins > LiteSpeed Cache"
        return 0
    fi

    info "Configuring WordPress LiteSpeed Cache Plugin Object Cache..."
    if bash "$plugin_script" --enable --connection-type=socket --redis-socket="$REDIS_SOCKET" --wp-root="$WP_ROOT"; then
        success "WordPress Object Cache configured successfully with Redis socket."
    else
        warning "Object Cache config script failed (non-fatal). Run manually:"
        warning "  bash $plugin_script --enable --wp-root=$WP_ROOT"
    fi
}

# ════════════════════════════════════════════════════════════════════════════
# MAIN
# ════════════════════════════════════════════════════════════════════════════

# Must be root
if [[ "$(id -u)" -ne 0 ]]; then
    error_exit "This script must be run as root."
fi

if [[ "$STATUS_ONLY" == "true" ]]; then
    do_status
    exit $?
fi

info "════════ Redis Install & Configure for WordPress (LLSMP/AlmaLinux) ════════"

# Run status check first — if Redis is already healthy, nothing to do
status_rc=0
do_status || status_rc=$?

if [[ $status_rc -eq 0 ]]; then
    success "Redis is already installed and running. No action needed."
    info "To reconfigure WordPress object cache, run:"
    info "  bash $SCRIPT_DIR/configure_litespeed_plugin_object_cache.sh --enable --wp-root=$WP_ROOT"
    exit 0
fi

if [[ $status_rc -eq 2 ]]; then
    # Installed but stopped — skip install/configure, just start it
    info "Redis is installed but stopped. Attempting to start..."
    step_start
    step_verify_access
    step_wp_object_cache
    success "════════ Redis re-enabled successfully ════════"
    exit 0
fi

# status_rc=1: not installed — full install flow
info "Redis not installed. Starting full installation..."
step_install
step_configure
step_socket_access
step_start
step_verify_access
step_wp_object_cache

success "════════ Redis setup complete ════════"
info "Socket : $REDIS_SOCKET"
info "Config : $(get_redis_conf)"
info "Test   : redis-cli -s $REDIS_SOCKET ping"