Make poll intervals and operational timeouts configurable

New env vars: UNLOAD_POLL_INTERVAL, HISTORY_POLL_INTERVAL, PROBE_TIMEOUT,
FREE_TIMEOUT, WARM_TIMEOUT, SHUTDOWN_TIMEOUT, PROMPT_CAPTURE_LIMIT.
Defaults unchanged; invalid values fail fast at startup.

cmd/gpu-turnstile/main.go

@@ -10,6 +10,7 @@ import (
 	"net/http"
 	"os"
 	"os/signal"
+	"strconv"
 	"strings"
 	"syscall"
 	"time"
@@ -32,9 +33,18 @@ type config struct {
 	unloadTimeout  time.Duration
 	jobTimeout     time.Duration
 	llmWaitTimeout time.Duration
-	warmModel      string
-	logLevel       slog.Level
-	logJSON        bool
+
+	unloadPollInterval  time.Duration
+	historyPollInterval time.Duration
+	probeTimeout        time.Duration
+	freeTimeout         time.Duration
+	warmTimeout         time.Duration
+	shutdownTimeout     time.Duration
+	promptCaptureLimit  int64
+
+	warmModel string
+	logLevel  slog.Level
+	logJSON   bool
 }
 
 func envDuration(getenv func(string) string, name string, dst *time.Duration) error {
@@ -59,7 +69,16 @@ func loadConfig(getenv func(string) string) (config, error) {
 		unloadTimeout:  time.Minute,
 		jobTimeout:     15 * time.Minute,
 		llmWaitTimeout: 10 * time.Minute,
-		logLevel:       slog.LevelInfo,
+
+		unloadPollInterval:  500 * time.Millisecond,
+		historyPollInterval: time.Second,
+		probeTimeout:        5 * time.Second,
+		freeTimeout:         30 * time.Second,
+		warmTimeout:         2 * time.Minute,
+		shutdownTimeout:     10 * time.Second,
+		promptCaptureLimit:  64 * 1024,
+
+		logLevel: slog.LevelInfo,
 	}
 	for _, e := range []struct {
 		name string
@@ -82,11 +101,24 @@ func loadConfig(getenv func(string) string) (config, error) {
 		{"UNLOAD_TIMEOUT", &cfg.unloadTimeout},
 		{"JOB_TIMEOUT", &cfg.jobTimeout},
 		{"LLM_WAIT_TIMEOUT", &cfg.llmWaitTimeout},
+		{"UNLOAD_POLL_INTERVAL", &cfg.unloadPollInterval},
+		{"HISTORY_POLL_INTERVAL", &cfg.historyPollInterval},
+		{"PROBE_TIMEOUT", &cfg.probeTimeout},
+		{"FREE_TIMEOUT", &cfg.freeTimeout},
+		{"WARM_TIMEOUT", &cfg.warmTimeout},
+		{"SHUTDOWN_TIMEOUT", &cfg.shutdownTimeout},
 	} {
 		if err := envDuration(getenv, e.name, e.dst); err != nil {
 			return cfg, err
 		}
 	}
+	if v := getenv("PROMPT_CAPTURE_LIMIT"); v != "" {
+		n, err := strconv.ParseInt(v, 10, 64)
+		if err != nil || n < 0 {
+			return cfg, fmt.Errorf("PROMPT_CAPTURE_LIMIT: must be a non-negative integer (bytes)")
+		}
+		cfg.promptCaptureLimit = n
+	}
 	if v := getenv("LOG_LEVEL"); v != "" {
 		var level slog.Level
 		if err := level.UnmarshalText([]byte(v)); err != nil {
@@ -139,17 +171,22 @@ func main() {
 	}
 
 	srv, err := proxy.New(proxy.Config{
-		OllamaURL:      cfg.ollamaURL,
-		ComfyURL:       cfg.comfyURL,
-		Lock:           lock.New(log),
-		Ollama:         ollamaClient,
-		Comfy:          comfyClient,
-		Metrics:        metrics.New(),
-		Log:            log,
-		LLMWaitTimeout: cfg.llmWaitTimeout,
-		UnloadTimeout:  cfg.unloadTimeout,
-		JobTimeout:     cfg.jobTimeout,
-		WarmModel:      cfg.warmModel,
+		OllamaURL:           cfg.ollamaURL,
+		ComfyURL:            cfg.comfyURL,
+		Lock:                lock.New(log),
+		Ollama:              ollamaClient,
+		Comfy:               comfyClient,
+		Metrics:             metrics.New(),
+		Log:                 log,
+		LLMWaitTimeout:      cfg.llmWaitTimeout,
+		UnloadTimeout:       cfg.unloadTimeout,
+		JobTimeout:          cfg.jobTimeout,
+		UnloadPollInterval:  cfg.unloadPollInterval,
+		HistoryPollInterval: cfg.historyPollInterval,
+		FreeTimeout:         cfg.freeTimeout,
+		WarmTimeout:         cfg.warmTimeout,
+		PromptCaptureLimit:  cfg.promptCaptureLimit,
+		WarmModel:           cfg.warmModel,
 	})
 	if err != nil {
 		log.Error("invalid configuration", "err", err)
@@ -157,7 +194,7 @@ func main() {
 	}
 
 	// Probe both upstreams once; failure is logged, not fatal.
-	probeCtx, probeCancel := context.WithTimeout(context.Background(), 5*time.Second)
+	probeCtx, probeCancel := context.WithTimeout(context.Background(), cfg.probeTimeout)
 	if err := ollamaClient.Probe(probeCtx); err != nil {
 		log.Warn("ollama probe failed", "url", cfg.ollamaURL, "err", err)
 	}
@@ -186,7 +223,7 @@ func main() {
 		log.Info("shutting down")
 	}
 
-	shutdownCtx, shutdownCancel := context.WithTimeout(context.Background(), 10*time.Second)
+	shutdownCtx, shutdownCancel := context.WithTimeout(context.Background(), cfg.shutdownTimeout)
 	defer shutdownCancel()
 	ollamaSrv.Shutdown(shutdownCtx)
 	comfySrv.Shutdown(shutdownCtx)