package main import ( "context" "flag" "log" "os/signal" "syscall" "time" "git.govekar.net/LovroG05/nvllama-watchdog/config" "git.govekar.net/LovroG05/nvllama-watchdog/core" "git.govekar.net/LovroG05/nvllama-watchdog/exporters" "git.govekar.net/LovroG05/nvllama-watchdog/readers" ) func main() { cfgPath := flag.String("config", "config/config.yaml", "path to the YAML config file") flag.Parse() cfg, err := config.Load(*cfgPath) if err != nil { log.Fatalf("config: %v", err) } // Build readers var rs []core.Reader for _, rc := range cfg.Readers { switch rc.Type { case "llamacpp": rs = append(rs, readers.NewLLamaCPPReader(rc.Endpoint, rc.Model)) case "nvidia_smi": rs = append(rs, readers.NewNvidiaSmiReader()) case "system": rs = append(rs, readers.NewSystemReader(rc.Enabled)) } } // Build exporters var es []core.Exporter var prom *exporters.PrometheusExporter for _, ec := range cfg.Exporters { switch ec.Type { case "prometheus": prom = exporters.NewPrometheusExporter(ec.Addr) es = append(es, prom) case "dev": es = append(es, exporters.NewDevExporter(ec.JSONFile)) case "influxdb": es = append(es, exporters.NewInfluxDBExporter(ec.URL, ec.Token, ec.User, ec.Pass)) } } // Start Prometheus HTTP listener in background if prom != nil { go func() { log.Printf("prometheus exporter listening on %s", prom.Addr()) if err := prom.Start(); err != nil { log.Fatalf("prometheus: %v", err) } }() } // Scrape loop ctx, stop := signal.NotifyContext(context.Background(), syscall.SIGINT, syscall.SIGTERM) defer stop() ticker := time.NewTicker(cfg.ScrapeInterval) defer ticker.Stop() for { select { case <-ctx.Done(): return case <-ticker.C: var all []core.Metric for _, r := range rs { ms, err := r.Read(ctx) if err != nil { log.Printf("reader %s: %v", r.Name(), err) continue } all = append(all, ms...) } for _, e := range es { if err := e.Write(ctx, all); err != nil { log.Printf("exporter %s: %v", e.Name(), err) } } } } }