a3afb2aeae
Model A prototype (run start.sh AS the manager, escalate only via helpers): - check_root.sh: accept the manager user, not root-only (init.sh keeps its own install-time root check). - init.sh: guard the top-level root-check + installer entrypoint with BASH_SOURCE!=$0 so it runs ONLY when init.sh is executed directly; when start.sh sources it as the manager the entrypoint (and its root check) no longer fires. Also: convert bare daemon-touching 'docker' calls (no helper -> hit the nonexistent /var/run socket in rootless) to runFileOp docker across app_status, app_health_*, network_prune, ip_is_available, check_docker_network, backup_db (db dumps) and crontab_check_processor. cd&&compose rooted-branches and 'docker compose --version' checks left as-is (rooted-only / no daemon). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> Signed-off-by: librelad <librelad@digitalangels.vip>
762 lines
22 KiB
Bash
Executable File
762 lines
22 KiB
Bash
Executable File
#!/bin/bash
|
|
|
|
# ========================================
|
|
# LIBREPORTAL SYSTEM HEALTH CHECKER
|
|
# Comprehensive system validation and repair tool
|
|
# ========================================
|
|
|
|
script_check_processor_flag="$1"
|
|
|
|
# Only run when executed directly, not when sourced
|
|
if [[ "$script_check_processor_flag" == "start_script" ]]; then
|
|
|
|
# --- Load the privilege helpers + docker-type config -------------------------
|
|
# cron runs this standalone, so the de-sudo helpers (runFileOp/runFileWrite/
|
|
# runSystem) and the config they key off are NOT otherwise in scope — without
|
|
# them every privileged op is "command not found". Same bootstrap as
|
|
# crontab_task_processor.sh. These files are pure function/var defs, safe to
|
|
# source.
|
|
LP_SCRIPTS="${install_scripts_dir:-/docker/install/scripts/}"
|
|
LP_DOCKER_CFG="/docker/configs/general/general_docker_install"
|
|
[[ -f "$LP_DOCKER_CFG" ]] && \
|
|
eval "$(grep -E '^CFG_DOCKER_INSTALL_(TYPE|USER)=' "$LP_DOCKER_CFG" | sed 's/[[:space:]]*#.*//')"
|
|
: "${sudo_user_name:=libreportal}"
|
|
: "${containers_dir:=/docker/containers/}"
|
|
: "${docker_dir:=/docker}"
|
|
for _lp_f in docker/command/run_privileged.sh \
|
|
docker/command/docker_run_install.sh \
|
|
checks/requirements/check_install_type.sh; do
|
|
[[ -f "${LP_SCRIPTS}${_lp_f}" ]] && source "${LP_SCRIPTS}${_lp_f}"
|
|
done
|
|
command -v resolveDockerInstallUser >/dev/null 2>&1 && resolveDockerInstallUser
|
|
|
|
# ========================================
|
|
# HEALTH CHECK CONFIGURATION
|
|
# ========================================
|
|
|
|
# Essential configuration (keeping your original structure)
|
|
TASK_DIR=""
|
|
if [[ "$TASK_DIR" == "" ]]; then
|
|
TASK_DIR="/docker/containers/libreportal/frontend/data/tasks"
|
|
fi
|
|
LOCK_FILE="$TASK_DIR/task_processor.lock"
|
|
|
|
# Health thresholds and limits
|
|
CRITICAL_FILES=("queue.json" "current.json")
|
|
REQUIRED_DIRECTORIES=("tasks")
|
|
SERVICE_NAME="libreportal"
|
|
MIN_DISK_SPACE_KB=10240 # 10MB minimum
|
|
MAX_LOG_SIZE_MB=50 # 50MB max log size
|
|
HEALTH_CHECK_TIMEOUT=30 # 30 seconds max per check
|
|
|
|
# Health status codes
|
|
HEALTH_STATUS_HEALTHY=0
|
|
HEALTH_STATUS_WARNING=1
|
|
HEALTH_STATUS_CRITICAL=2
|
|
HEALTH_STATUS_ERROR=3
|
|
|
|
# Logging
|
|
HEALTH_LOG_FILE="$TASK_DIR/health_check.log"
|
|
HEALTH_REPORT_FILE="$TASK_DIR/health_report.json"
|
|
|
|
# ========================================
|
|
# HEALTH LOGGING SYSTEM
|
|
# ========================================
|
|
|
|
healthLogInfo() {
|
|
local message="$1"
|
|
local timestamp=$(date '+%Y-%m-%d %H:%M:%S')
|
|
echo "$timestamp [INFO] [HEALTH_CHECK] $message" | tee -a "$HEALTH_LOG_FILE"
|
|
}
|
|
|
|
healthLogError() {
|
|
local message="$1"
|
|
local timestamp=$(date '+%Y-%m-%d %H:%M:%S')
|
|
echo "$timestamp [ERROR] [HEALTH_CHECK] $message" | tee -a "$HEALTH_LOG_FILE"
|
|
}
|
|
|
|
healthLogWarning() {
|
|
local message="$1"
|
|
local timestamp=$(date '+%Y-%m-%d %H:%M:%S')
|
|
echo "$timestamp [WARN] [HEALTH_CHECK] $message" | tee -a "$HEALTH_LOG_FILE"
|
|
}
|
|
|
|
healthLogSuccess() {
|
|
local message="$1"
|
|
local timestamp=$(date '+%Y-%m-%d %H:%M:%S')
|
|
echo "$timestamp [SUCCESS] [HEALTH_CHECK] $message" | tee -a "$HEALTH_LOG_FILE"
|
|
}
|
|
|
|
# ========================================
|
|
# COMPREHENSIVE SYSTEM VALIDATION
|
|
# ========================================
|
|
|
|
validateSystemHealth() {
|
|
healthLogInfo "Starting comprehensive system health check..."
|
|
|
|
local issues=()
|
|
local warnings=()
|
|
local critical_issues=()
|
|
|
|
# 1. LibrePortal Installation Check
|
|
healthLogInfo "Checking LibrePortal installation..."
|
|
if ! validateLibrePortalInstallation; then
|
|
critical_issues+=("libreportal_not_installed")
|
|
fi
|
|
|
|
# 2. Docker Service Check
|
|
healthLogInfo "Checking Docker service..."
|
|
if ! validateDockerService; then
|
|
critical_issues+=("docker_service_not_running")
|
|
fi
|
|
|
|
# 3. Container Health Check
|
|
healthLogInfo "Checking LibrePortal container..."
|
|
if ! validateContainerHealth; then
|
|
issues+=("container_health_issues")
|
|
fi
|
|
|
|
# 4. WebUI Readiness Check
|
|
healthLogInfo "Checking WebUI readiness..."
|
|
if ! validateWebUIReadiness; then
|
|
issues+=("webui_not_ready")
|
|
fi
|
|
|
|
# 5. Directory Structure Check
|
|
healthLogInfo "Checking directory structure..."
|
|
if ! validateDirectoryStructure; then
|
|
warnings+=("directory_structure_issues")
|
|
fi
|
|
|
|
# 6. File System Check
|
|
healthLogInfo "Checking file system..."
|
|
if ! validateFileSystem; then
|
|
issues+=("file_system_issues")
|
|
fi
|
|
|
|
# 7. Permissions Check
|
|
healthLogInfo "Checking file permissions..."
|
|
if ! validatePermissions; then
|
|
issues+=("permission_issues")
|
|
fi
|
|
|
|
# 8. Disk Space Check
|
|
healthLogInfo "Checking disk space..."
|
|
if ! validateDiskSpace; then
|
|
critical_issues+=("insufficient_disk_space")
|
|
fi
|
|
|
|
# 9. Task System Check
|
|
healthLogInfo "Checking task system..."
|
|
if ! validateTaskSystem; then
|
|
issues+=("task_system_issues")
|
|
fi
|
|
|
|
# 10. Service Status Check
|
|
healthLogInfo "Checking systemd service..."
|
|
if ! validateSystemService; then
|
|
warnings+=("service_not_optimal")
|
|
fi
|
|
|
|
# 11. Log Health Check
|
|
healthLogInfo "Checking log health..."
|
|
if ! validateLogHealth; then
|
|
warnings+=("log_health_issues")
|
|
fi
|
|
|
|
# Generate results
|
|
local total_issues=$((${#issues[@]} + ${#warnings[@]} + ${#critical_issues[@]}))
|
|
|
|
if [ ${#critical_issues[@]} -gt 0 ]; then
|
|
healthLogError "❌ CRITICAL ISSUES FOUND: ${critical_issues[*]}"
|
|
|
|
elif [ ${#issues[@]} -gt 0 ]; then
|
|
healthLogWarning "⚠️ ISSUES FOUND: ${issues[*]}"
|
|
|
|
elif [ ${#warnings[@]} -gt 0 ]; then
|
|
healthLogWarning "⚠️ WARNINGS: ${warnings[*]}"
|
|
|
|
else
|
|
healthLogSuccess "✅ All health checks passed"
|
|
|
|
fi
|
|
}
|
|
|
|
validateLibrePortalInstallation() {
|
|
if command -v libreportal >/dev/null 2>&1; then
|
|
local version=$(libreportal --version 2>/dev/null)
|
|
healthLogSuccess "LibrePortal installed: $version"
|
|
else
|
|
healthLogError "LibrePortal command not found"
|
|
fi
|
|
}
|
|
|
|
validateDockerService() {
|
|
if command -v docker >/dev/null 2>&1; then
|
|
if systemctl is-active --quiet docker; then
|
|
healthLogSuccess "Docker service is running"
|
|
else
|
|
healthLogError "Docker service is not running"
|
|
fi
|
|
else
|
|
healthLogError "Docker command not found"
|
|
fi
|
|
}
|
|
|
|
validateContainerHealth() {
|
|
local containerInfo=$(runFileOp docker ps -a --filter "name=libreportal-service" --format "{{.Status}}|{{.Names}}" 2>/dev/null)
|
|
|
|
if [ -z "$containerInfo" ]; then
|
|
healthLogError "LibrePortal WebUI container not found"
|
|
fi
|
|
|
|
local status=$(echo "$containerInfo" | cut -d'|' -f1)
|
|
|
|
if echo "$status" | grep -q "Up"; then
|
|
# Check if container is responsive
|
|
if runFileOp docker exec libreportal-service pgrep -f "node.*webui" >/dev/null 2>&1; then
|
|
healthLogSuccess "Container is running and responsive"
|
|
else
|
|
healthLogWarning "Container is running but WebUI process not found"
|
|
fi
|
|
else
|
|
healthLogWarning "Container status: $status"
|
|
fi
|
|
}
|
|
|
|
validateWebUIReadiness() {
|
|
# Get the actual port mapping from the container
|
|
local portMapping=$(runFileOp docker port libreportal-service 2>/dev/null | head -1)
|
|
|
|
if [ -z "$portMapping" ]; then
|
|
healthLogWarning "No port mapping found for libreportal-service container"
|
|
fi
|
|
|
|
# Extract the host port
|
|
local hostPort=$(echo "$portMapping" | sed 's/.*://' | sed 's/[^0-9]//g')
|
|
|
|
if [ -z "$hostPort" ]; then
|
|
healthLogWarning "Could not extract host port from: $portMapping"
|
|
fi
|
|
|
|
healthLogInfo "Checking WebUI on port: $hostPort"
|
|
|
|
# Try multiple endpoints to check if WebUI is ready
|
|
local endpoints=(
|
|
"/" # Main page
|
|
"/api/health" # Health endpoint (if exists)
|
|
"/js/system-loader.js" # A known JavaScript file
|
|
"/css/main.css" # A known CSS file
|
|
)
|
|
|
|
for endpoint in "${endpoints[@]}"; do
|
|
if curl -s --max-time 3 "http://localhost:$hostPort$endpoint" >/dev/null 2>&1; then
|
|
healthLogInfo "WebUI is responding on port $hostPort (endpoint: $endpoint)"
|
|
|
|
# Additional check: verify it's actually the LibrePortal WebUI
|
|
if curl -s --max-time 3 "http://localhost:$hostPort/" | grep -q -i "libreportal\|webui\|dashboard" 2>/dev/null; then
|
|
healthLogSuccess "WebUI is ready on port $hostPort"
|
|
else
|
|
healthLogWarning "Port $hostPort is responding but doesn't appear to be LibrePortal WebUI"
|
|
fi
|
|
fi
|
|
done
|
|
|
|
healthLogWarning "WebUI not ready on port $hostPort"
|
|
}
|
|
|
|
validateDirectoryStructure() {
|
|
local issues=()
|
|
|
|
# Check main tasks directory
|
|
if [ ! -d "$TASK_DIR" ]; then
|
|
issues+=("tasks_directory_missing")
|
|
fi
|
|
|
|
# Check subdirectories
|
|
for dir in "${REQUIRED_DIRECTORIES[@]}"; do
|
|
local dirPath="$TASK_DIR/$dir"
|
|
if [ ! -d "$dirPath" ]; then
|
|
issues+=("directory_missing:$dir")
|
|
fi
|
|
done
|
|
|
|
if [ ${#issues[@]} -gt 0 ]; then
|
|
healthLogWarning "Directory issues: ${issues[*]}"
|
|
else
|
|
healthLogSuccess "Directory structure is valid"
|
|
fi
|
|
}
|
|
|
|
validateFileSystem() {
|
|
local issues=()
|
|
|
|
# Check critical files
|
|
for file in "${CRITICAL_FILES[@]}"; do
|
|
local filePath="$TASK_DIR/$file"
|
|
|
|
if [ ! -f "$filePath" ]; then
|
|
issues+=("missing_file:$file")
|
|
elif [ ! -r "$filePath" ]; then
|
|
issues+=("file_not_readable:$file")
|
|
elif [ ! -w "$filePath" ]; then
|
|
issues+=("file_not_writable:$file")
|
|
fi
|
|
done
|
|
|
|
if [ ${#issues[@]} -gt 0 ]; then
|
|
healthLogWarning "File system issues: ${issues[*]}"
|
|
else
|
|
healthLogSuccess "File system is valid"
|
|
fi
|
|
}
|
|
|
|
validatePermissions() {
|
|
local issues=()
|
|
|
|
# Check directory permissions
|
|
if [ ! -w "$TASK_DIR" ]; then
|
|
issues+=("directory_not_writable")
|
|
fi
|
|
|
|
# Check file permissions for critical files
|
|
for file in "${CRITICAL_FILES[@]}"; do
|
|
local filePath="$TASK_DIR/$file"
|
|
if [ -f "$filePath" ]; then
|
|
if [ ! -r "$filePath" ] || [ ! -w "$filePath" ]; then
|
|
issues+=("permission_issues:$file")
|
|
fi
|
|
fi
|
|
done
|
|
|
|
if [ ${#issues[@]} -gt 0 ]; then
|
|
healthLogWarning "Permission issues: ${issues[*]}"
|
|
else
|
|
healthLogSuccess "Permissions are valid"
|
|
fi
|
|
}
|
|
|
|
validateDiskSpace() {
|
|
local availableSpace=$(df "$TASK_DIR" 2>/dev/null | awk 'NR==2 {print $4}')
|
|
|
|
if [ -z "$availableSpace" ]; then
|
|
healthLogError "Could not determine disk space"
|
|
fi
|
|
|
|
if [ "$availableSpace" -lt $MIN_DISK_SPACE_KB ]; then
|
|
healthLogError "Insufficient disk space: ${availableSpace}KB available, ${MIN_DISK_SPACE_KB}KB required"
|
|
else
|
|
healthLogSuccess "Sufficient disk space: ${availableSpace}KB available"
|
|
fi
|
|
}
|
|
|
|
validateTaskSystem() {
|
|
local issues=()
|
|
|
|
# Validate queue.json format
|
|
if [ -f "$TASK_DIR/queue.json" ]; then
|
|
if ! python3 -c "import json; json.load(open('$TASK_DIR/queue.json'))" 2>/dev/null; then
|
|
issues+=("queue_json_invalid")
|
|
fi
|
|
fi
|
|
|
|
# Validate current.json format
|
|
if [ -f "$TASK_DIR/current.json" ]; then
|
|
if ! python3 -c "import json; json.load(open('$TASK_DIR/current.json'))" 2>/dev/null; then
|
|
issues+=("current_json_invalid")
|
|
fi
|
|
fi
|
|
|
|
# Check for orphaned task files
|
|
local queueContent=""
|
|
local currentContent=""
|
|
|
|
if [ -f "$TASK_DIR/queue.json" ]; then
|
|
queueContent=$(cat "$TASK_DIR/queue.json" 2>/dev/null)
|
|
fi
|
|
|
|
if [ -f "$TASK_DIR/current.json" ]; then
|
|
currentContent=$(cat "$TASK_DIR/current.json" 2>/dev/null)
|
|
fi
|
|
|
|
# Find task files not referenced in queue or current
|
|
for taskFile in "$TASK_DIR"/task_*.json; do
|
|
if [ -f "$taskFile" ]; then
|
|
local taskId=$(basename "$taskFile" .json)
|
|
if ! echo "$queueContent $currentContent" | grep -q "$taskId"; then
|
|
issues+=("orphaned_task_file:$taskId")
|
|
fi
|
|
fi
|
|
done
|
|
|
|
if [ ${#issues[@]} -gt 0 ]; then
|
|
healthLogWarning "Task system issues: ${issues[*]}"
|
|
else
|
|
healthLogSuccess "Task system is valid"
|
|
fi
|
|
}
|
|
|
|
validateSystemService() {
|
|
if systemctl is-enabled --quiet "$SERVICE_NAME" 2>/dev/null; then
|
|
if systemctl is-active --quiet "$SERVICE_NAME" 2>/dev/null; then
|
|
healthLogSuccess "Systemd service is enabled and active"
|
|
else
|
|
healthLogWarning "Systemd service is enabled but not active"
|
|
fi
|
|
else
|
|
healthLogWarning "Systemd service is not enabled"
|
|
fi
|
|
}
|
|
|
|
validateLogHealth() {
|
|
local issues=()
|
|
|
|
# Check log file sizes
|
|
for logFile in "$TASK_DIR"/*.log; do
|
|
if [ -f "$logFile" ]; then
|
|
local fileSize=$(stat -f%z "$logFile" 2>/dev/null || stat -c%s "$logFile" 2>/dev/null)
|
|
if [ -n "$fileSize" ]; then
|
|
local fileSizeMB=$((fileSize / 1024 / 1024))
|
|
if [ $fileSizeMB -gt $MAX_LOG_SIZE_MB ]; then
|
|
issues+=("large_log_file:$(basename $logFile):${fileSizeMB}MB")
|
|
fi
|
|
fi
|
|
fi
|
|
done
|
|
|
|
if [ ${#issues[@]} -gt 0 ]; then
|
|
healthLogWarning "Log health issues: ${issues[*]}"
|
|
else
|
|
healthLogSuccess "Log files are healthy"
|
|
fi
|
|
}
|
|
|
|
# ========================================
|
|
# ADVANCED REPAIR TOOLS
|
|
# ========================================
|
|
|
|
repairSystemIssues() {
|
|
healthLogInfo "Starting system repair process..."
|
|
|
|
local repaired=0
|
|
local failed=0
|
|
|
|
# 1. Repair directory structure
|
|
if repairDirectoryStructure; then
|
|
((repaired++))
|
|
else
|
|
((failed++))
|
|
fi
|
|
|
|
# 2. Repair file system
|
|
if repairFileSystem; then
|
|
((repaired++))
|
|
else
|
|
((failed++))
|
|
fi
|
|
|
|
# 3. Repair permissions
|
|
if repairPermissions; then
|
|
((repaired++))
|
|
else
|
|
((failed++))
|
|
fi
|
|
|
|
# 4. Repair task system
|
|
if repairTaskSystem; then
|
|
((repaired++))
|
|
else
|
|
((failed++))
|
|
fi
|
|
|
|
# 5. Repair service
|
|
if repairSystemService; then
|
|
((repaired++))
|
|
else
|
|
((failed++))
|
|
fi
|
|
|
|
healthLogInfo "Repair completed: $repaired repaired, $failed failed"
|
|
|
|
if [ $failed -eq 0 ]; then
|
|
healthLogSuccess "All repairs completed successfully"
|
|
else
|
|
healthLogWarning "Some repairs failed"
|
|
fi
|
|
}
|
|
|
|
repairDirectoryStructure() {
|
|
healthLogInfo "Repairing directory structure..."
|
|
|
|
# Create main tasks directory
|
|
if [ ! -d "$TASK_DIR" ]; then
|
|
healthLogInfo "Creating tasks directory: $TASK_DIR"
|
|
runFileOp mkdir -p "$TASK_DIR" || {
|
|
healthLogError "Failed to create tasks directory"
|
|
exit 1
|
|
}
|
|
fi
|
|
|
|
# Create required subdirectories
|
|
for dir in "${REQUIRED_DIRECTORIES[@]}"; do
|
|
local dirPath="$TASK_DIR/$dir"
|
|
if [ ! -d "$dirPath" ]; then
|
|
healthLogInfo "Creating directory: $dirPath"
|
|
runFileOp mkdir -p "$dirPath" || {
|
|
healthLogError "Failed to create directory: $dirPath"
|
|
exit 1
|
|
}
|
|
fi
|
|
done
|
|
|
|
healthLogSuccess "Directory structure repaired"
|
|
}
|
|
|
|
repairFileSystem() {
|
|
healthLogInfo "Repairing file system..."
|
|
|
|
local repaired=0
|
|
|
|
# Create missing critical files
|
|
for file in "${CRITICAL_FILES[@]}"; do
|
|
local filePath="$TASK_DIR/$file"
|
|
|
|
if [ ! -f "$filePath" ]; then
|
|
healthLogInfo "Creating missing file: $file"
|
|
case "$file" in
|
|
"queue.json")
|
|
echo "[]" | runFileWrite "$filePath" && ((repaired++))
|
|
;;
|
|
"current.json")
|
|
echo "{}" | runFileWrite "$filePath" && ((repaired++))
|
|
;;
|
|
esac
|
|
fi
|
|
done
|
|
|
|
healthLogSuccess "File system repaired: $repaired files created"
|
|
}
|
|
|
|
repairPermissions() {
|
|
healthLogInfo "Repairing permissions..."
|
|
|
|
# Fix directory permissions
|
|
runFileOp chmod 755 "$TASK_DIR" 2>/dev/null
|
|
|
|
# Fix file permissions
|
|
for file in "${CRITICAL_FILES[@]}"; do
|
|
local filePath="$TASK_DIR/$file"
|
|
if [ -f "$filePath" ]; then
|
|
runFileOp chmod 644 "$filePath" 2>/dev/null
|
|
fi
|
|
done
|
|
|
|
# Set ownership (if docker_install_user is available)
|
|
if [ -n "$docker_install_user" ]; then
|
|
runFileOp chown -R $docker_install_user:$docker_install_user "$TASK_DIR" 2>/dev/null
|
|
fi
|
|
|
|
healthLogSuccess "Permissions repaired"
|
|
}
|
|
|
|
repairTaskSystem() {
|
|
healthLogInfo "Repairing task system..."
|
|
|
|
local repaired=0
|
|
|
|
# Validate and repair queue.json
|
|
if [ -f "$TASK_DIR/queue.json" ]; then
|
|
if ! python3 -c "import json; json.load(open('$TASK_DIR/queue.json'))" 2>/dev/null; then
|
|
healthLogInfo "Repairing corrupted queue.json"
|
|
echo "[]" | runFileWrite "$TASK_DIR/queue.json" && ((repaired++))
|
|
fi
|
|
fi
|
|
|
|
# Validate and repair current.json
|
|
if [ -f "$TASK_DIR/current.json" ]; then
|
|
if ! python3 -c "import json; json.load(open('$TASK_DIR/current.json'))" 2>/dev/null; then
|
|
healthLogInfo "Repairing corrupted current.json"
|
|
echo "{}" | runFileWrite "$TASK_DIR/current.json" && ((repaired++))
|
|
fi
|
|
fi
|
|
|
|
# Clean up orphaned task files
|
|
local queueContent=""
|
|
local currentContent=""
|
|
|
|
if [ -f "$TASK_DIR/queue.json" ]; then
|
|
queueContent=$(cat "$TASK_DIR/queue.json" 2>/dev/null)
|
|
fi
|
|
|
|
if [ -f "$TASK_DIR/current.json" ]; then
|
|
currentContent=$(cat "$TASK_DIR/current.json" 2>/dev/null)
|
|
fi
|
|
|
|
for taskFile in "$TASK_DIR"/task_*.json; do
|
|
if [ -f "$taskFile" ]; then
|
|
local taskId=$(basename "$taskFile" .json)
|
|
if ! echo "$queueContent $currentContent" | grep -q "$taskId"; then
|
|
healthLogInfo "Removing orphaned task file: $taskId"
|
|
runFileOp rm -f "$taskFile" && ((repaired++))
|
|
fi
|
|
fi
|
|
done
|
|
|
|
healthLogSuccess "Task system repaired: $repaired issues fixed"
|
|
}
|
|
|
|
repairSystemService() {
|
|
healthLogInfo "Repairing systemd service..."
|
|
|
|
# Enable service if not enabled
|
|
if ! systemctl is-enabled --quiet "$SERVICE_NAME" 2>/dev/null; then
|
|
healthLogInfo "Enabling systemd service: $SERVICE_NAME"
|
|
runSystem systemctl enable "$SERVICE_NAME" 2>/dev/null
|
|
fi
|
|
|
|
# Start service if not active
|
|
if ! systemctl is-active --quiet "$SERVICE_NAME" 2>/dev/null; then
|
|
healthLogInfo "Starting systemd service: $SERVICE_NAME"
|
|
runSystem systemctl start "$SERVICE_NAME" 2>/dev/null
|
|
fi
|
|
|
|
healthLogSuccess "Systemd service repaired"
|
|
}
|
|
|
|
# ========================================
|
|
# HEALTH REPORTING
|
|
# ========================================
|
|
|
|
generateHealthReport() {
|
|
healthLogInfo "Generating comprehensive health report..."
|
|
|
|
local report="{"
|
|
report+="\"timestamp\":\"$(date -Iseconds)\","
|
|
report+="\"status\":\"$1\","
|
|
report+="\"checks\":{"
|
|
|
|
# Add individual check results
|
|
report+="\"libreportal_installation\":$(validateLibrePortalInstallation && echo "true" || echo "false"),"
|
|
report+="\"docker_service\":$(validateDockerService && echo "true" || echo "false"),"
|
|
report+="\"container_health\":$(validateContainerHealth && echo "true" || echo "false"),"
|
|
report+="\"webui_readiness\":$(validateWebUIReadiness && echo "true" || echo "false"),"
|
|
report+="\"directory_structure\":$(validateDirectoryStructure && echo "true" || echo "false"),"
|
|
report+="\"file_system\":$(validateFileSystem && echo "true" || echo "false"),"
|
|
report+="\"permissions\":$(validatePermissions && echo "true" || echo "false"),"
|
|
report+="\"disk_space\":$(validateDiskSpace && echo "true" || echo "false"),"
|
|
report+="\"task_system\":$(validateTaskSystem && echo "true" || echo "false"),"
|
|
report+="\"system_service\":$(validateSystemService && echo "true" || echo "false"),"
|
|
report+="\"log_health\":$(validateLogHealth && echo "true" || echo "false")"
|
|
|
|
report+="},"
|
|
report+="\"system_info\":{"
|
|
report+="\"task_dir\":\"$TASK_DIR\","
|
|
report+="\"service_name\":\"$SERVICE_NAME\","
|
|
report+="\"disk_available_kb\":\"$(df \"$TASK_DIR\" 2>/dev/null | awk 'NR==2 {print $4}')\""
|
|
report+="}"
|
|
report+="}"
|
|
|
|
echo "$report" | runFileWrite "$HEALTH_REPORT_FILE"
|
|
healthLogSuccess "Health report generated: $HEALTH_REPORT_FILE"
|
|
}
|
|
|
|
# ========================================
|
|
# MAINTENANCE OPERATIONS
|
|
# ========================================
|
|
|
|
performMaintenance() {
|
|
healthLogInfo "Performing system maintenance..."
|
|
|
|
# Clean up old logs
|
|
find "$TASK_DIR" -name "*.log" -mtime +7 -delete 2>/dev/null
|
|
|
|
# Clean up temporary files
|
|
find "$TASK_DIR" -name "*.tmp" -delete 2>/dev/null
|
|
find "$TASK_DIR" -name ".queue.*" -delete 2>/dev/null
|
|
|
|
# Rotate large logs
|
|
for logFile in "$TASK_DIR"/*.log; do
|
|
if [ -f "$logFile" ]; then
|
|
local fileSize=$(stat -f%z "$logFile" 2>/dev/null || stat -c%s "$logFile" 2>/dev/null)
|
|
if [ -n "$fileSize" ]; then
|
|
local fileSizeMB=$((fileSize / 1024 / 1024))
|
|
if [ $fileSizeMB -gt $MAX_LOG_SIZE_MB ]; then
|
|
healthLogInfo "Rotating large log: $(basename $logFile) (${fileSizeMB}MB)"
|
|
runFileOp mv "$logFile" "${logFile}.old"
|
|
runFileOp touch "$logFile"
|
|
fi
|
|
fi
|
|
fi
|
|
done
|
|
|
|
healthLogSuccess "Maintenance completed"
|
|
}
|
|
|
|
# ========================================
|
|
# MAIN HEALTH CHECK FUNCTION
|
|
# ========================================
|
|
|
|
check_task_processor_health() {
|
|
healthLogInfo "=== LibrePortal System Health Check Starting ==="
|
|
|
|
# Run comprehensive health check
|
|
validateSystemHealth
|
|
local healthStatus=$?
|
|
|
|
# Generate health report
|
|
case $healthStatus in
|
|
$HEALTH_STATUS_HEALTHY)
|
|
generateHealthReport "healthy"
|
|
healthLogSuccess "✅ System is healthy"
|
|
;;
|
|
$HEALTH_STATUS_WARNING)
|
|
generateHealthReport "warning"
|
|
healthLogWarning "⚠️ System has warnings"
|
|
;;
|
|
$HEALTH_STATUS_CRITICAL)
|
|
generateHealthReport "critical"
|
|
healthLogError "❌ System has critical issues"
|
|
;;
|
|
$HEALTH_STATUS_ERROR)
|
|
generateHealthReport "error"
|
|
healthLogError "❌ Health check failed with errors"
|
|
;;
|
|
esac
|
|
|
|
healthLogInfo "=== Health Check Complete ==="
|
|
|
|
}
|
|
|
|
# ========================================
|
|
# COMMAND LINE INTERFACE
|
|
# ========================================
|
|
|
|
case "$script_check_processor_flag" in
|
|
"check")
|
|
check_task_processor_health
|
|
;;
|
|
"repair")
|
|
healthLogInfo "Starting system repair..."
|
|
repairSystemIssues
|
|
;;
|
|
"report")
|
|
generateHealthReport "manual"
|
|
echo "Health report generated: $HEALTH_REPORT_FILE"
|
|
;;
|
|
"maintenance")
|
|
performMaintenance
|
|
;;
|
|
"start_script")
|
|
check_task_processor_health
|
|
;;
|
|
*)
|
|
echo "Usage: $0 {check|repair|report|maintenance|start_script}"
|
|
echo " check - Run comprehensive health check"
|
|
echo " repair - Attempt to repair detected issues"
|
|
echo " report - Generate detailed health report"
|
|
echo " maintenance- Perform system maintenance"
|
|
echo " start_script- Run health check (for cron compatibility)"
|
|
exit 1
|
|
;;
|
|
esac
|
|
fi |