watch-tool/main.go

package main

import (
	"codeberg.org/pata1704/drain3"
	"context"
	"log/slog"
	"os"
	"os/signal"
	"sync"
	"syscall"
	"time"
	"watch-tool/helpers"
	"watch-tool/models"
	"watch-tool/patterns"
)

var currentHostname string

func init() {
	var err error
	currentHostname, err = os.Hostname()
	if err != nil {
		currentHostname = "unknown"
		slog.Warn("Could not determine hostname, using fallback", "fallback", currentHostname)
	}
}

func main() {
	cfg, err := LoadConfig()
	if err != nil {
		slog.Error("Startup failed: configuration error", "error", err)
		os.Exit(1)
	}

	slog.Info("System Monitor started", "hostname", currentHostname)

	if err := patterns.GetInstance().Load(cfg.PatternsFile); err != nil {
		slog.Error("Startup failed: could not load patterns", "file", cfg.PatternsFile, "error", err)
		os.Exit(1)
	}
	slog.Info("Regex patterns loaded successfully", "file", cfg.PatternsFile)

	var d3Cfg *drain3.Config
	if cfg.Drain3.Enabled {
		d3Cfg = &drain3.Config{
			Depth:       cfg.Drain3.Depth,
			SimTh:       cfg.Drain3.SimThreshold,
			MaxChildren: cfg.Drain3.MaxChildren,
		}
		slog.Info("Drain3 anomaly detection enabled", "state_dir", cfg.Drain3.StateDir)
	} else {
		slog.Info("Drain3 anomaly detection disabled")
	}

	var storage StorageInterface
	if cfg.LocalStorage.Enable {
		rotationConfig := StorageRotationConfig{
			MaxSizeBytes:         cfg.LocalStorage.RotationConfig.MaxSizeBytes,
			MaxAgeHours:          cfg.LocalStorage.RotationConfig.GetMaxAge(),
			MaxFiles:             cfg.LocalStorage.RotationConfig.MaxFiles,
			CheckIntervalMinutes: cfg.LocalStorage.RotationConfig.GetCheckInterval(),
			ArchiveDir:           cfg.LocalStorage.RotationConfig.ArchiveDir,
		}

		sqliteStorage, err := NewSQLiteStorageWithRotation(cfg.LocalStorage.DBPath, rotationConfig)
		if err != nil {
			slog.Error("failed to initialize SQLite storage", "error", err)
			os.Exit(1)
		}
		storage = sqliteStorage
		defer storage.Close()
		slog.Info("SQLite storage initialized", "path", cfg.LocalStorage.DBPath)
	} else {
		slog.Error("Local storage is disabled, but it's required for the new architecture")
		os.Exit(1)
	}

	var exportManager *ExportManager
	if cfg.Export.Enabled {
		exportConfig := ExportManagerConfig{
			BatchSize:           cfg.Export.BatchSize,
			ExportInterval:      cfg.Export.ExportInterval,
			RetryAttempts:       cfg.Export.RetryAttempts,
			RetryBackoff:        cfg.Export.RetryBackoff,
			HealthCheckInterval: cfg.Export.HealthCheckInterval,
		}

		exportManager = NewExportManager(storage, exportConfig)

		if cfg.Elasticsearch.Enabled {
			esExporter, err := NewElasticsearchExporter(cfg.Elasticsearch)
			if err != nil {
				slog.Error("failed to create Elasticsearch exporter", "error", err)
				os.Exit(1)
			}

			if err := esExporter.HealthCheck(context.Background()); err != nil {
				slog.Error("Elasticsearch health check failed", "error", err)
				os.Exit(1)
			}

			exportManager.RegisterExporter("elasticsearch", esExporter)
			slog.Info("Elasticsearch exporter registered")
		}
	}

	logChan := make(chan models.LogMessage, 1000)
	ctx, cancel := context.WithCancel(context.Background())
	defer cancel()

	var wg sync.WaitGroup

	wg.Add(1)
	helpers.SafeGo(ctx, "LogProcessor", func() {
		defer wg.Done()
		processor := NewLogProcessor(storage)
		processor.Start(ctx, logChan)
	})

	if exportManager != nil {
		wg.Add(1)
		helpers.SafeGo(ctx, "ExportManager", func() {
			defer wg.Done()
			exportManager.Start(ctx)
		})
	}

	for _, service := range cfg.Services {
		if !service.Enabled {
			slog.Debug("Service deactivated, skipping...", "service", service.Name)
			continue
		}

		wg.Add(1)
		srv := service

		helpers.SafeGo(ctx, "ServiceMonitor-"+srv.Name, func() {
			defer wg.Done()
			monitor := NewServiceMonitor(srv, currentHostname, d3Cfg, cfg.Drain3.StateDir)

			if err := monitor.Start(ctx, logChan); err != nil {
				slog.Error("Error watching service", "service", srv.Name, "error", err)
			}
		})

		slog.Info("Started watching Service-Log", "service", service.Name)
	}

	for _, tool := range cfg.Tools {
		if !tool.Enabled {
			slog.Debug("Tool is deactivated, skipping...", "tool", tool.Name)
			continue
		}

		wg.Add(1)
		t := tool

		helpers.SafeGo(ctx, "FileMonitor-"+t.Name, func() {
			defer wg.Done()

			monitor := NewFileMonitor(t, currentHostname, d3Cfg, cfg.Drain3.StateDir)

			if err := monitor.Start(ctx, logChan); err != nil {
				slog.Error("Error watching tool", "tool", t.Name, "error", err)
			}
		})

		slog.Info("Started watching logs", "tool", tool.Name, "file", tool.LogFile)
	}

	if cfg.SystemMetrics.Enabled {
		wg.Add(1)
		helpers.SafeGo(ctx, "SystemMetrics", func() {
			defer wg.Done()
			collector := NewSystemMetricsCollector(cfg.SystemMetrics, cfg.PollIntervalSeconds, currentHostname)
			collector.Start(ctx, storage, logChan)
		})
		slog.Info("Started collecting System-Metrics")
	}

	if cfg.WebService.Enabled {
		wg.Add(1)
		helpers.SafeGo(ctx, "WebService", func() {
			defer wg.Done()
			webService := NewWebService(cfg, storage)
			if err := webService.Start(ctx); err != nil {
				slog.Error("Web service error", "error", err)
			}
		})
		slog.Info("Web service started", "host", cfg.WebService.Host, "port", cfg.WebService.Port)
	}

	sigCh := make(chan os.Signal, 1)
	signal.Notify(sigCh, os.Interrupt, syscall.SIGTERM)

	s := <-sigCh
	slog.Info("Shutdown signal received, stopping threads...", "signal", s)

	cancel()

	done := make(chan struct{})
	go func() {
		wg.Wait()
		close(done)
	}()

	select {
	case <-done:
		close(logChan)
		slog.Info("All threads closed gracefully")
	case <-time.After(10 * time.Second):
		slog.Error("Shutdown timeout reached, force quitting")
		os.Exit(1)
	}

	slog.Info("Program stopped")
}