server: guard nil liveParams when no orchestrators available for AI request

CHANGELOG_PENDING.md

@@ -18,4 +18,6 @@
 
 #### General
 
+- Fix gateway nil-pointer panic on non-live AI requests when no orchestrators are available
+
 #### CLI

server/ai_process.go

@@ -1588,17 +1588,20 @@ func processAIRequest(ctx context.Context, params aiRequestParams, req interface
 		if monitor.Enabled {
 			monitor.AIRequestError(errMsg, monitor.ToPipeline(capName), modelID, nil)
 		}
-		monitor.SendQueueEventAsync("stream_trace", map[string]interface{}{
-			"type":        "gateway_no_orchestrators_available",
-			"timestamp":   time.Now().UnixMilli(),
-			"stream_id":   params.liveParams.streamID,
-			"pipeline_id": params.liveParams.pipelineID,
-			"request_id":  params.liveParams.requestID,
-			"orchestrator_info": map[string]interface{}{
-				"address": "",
-				"url":     "",
-			},
-		})
+		// liveParams is only set for realtime video requests; guard against nil for other pipelines
+		if params.liveParams != nil {
+			monitor.SendQueueEventAsync("stream_trace", map[string]interface{}{
+				"type":        "gateway_no_orchestrators_available",
+				"timestamp":   time.Now().UnixMilli(),
+				"stream_id":   params.liveParams.streamID,
+				"pipeline_id": params.liveParams.pipelineID,
+				"request_id":  params.liveParams.requestID,
+				"orchestrator_info": map[string]interface{}{
+					"address": "",
+					"url":     "",
+				},
+			})
+		}
 		return nil, &ServiceUnavailableError{err: errors.New(errMsg)}
 	}
 	return resp, nil

server/ai_process_test.go

@@ -4,9 +4,12 @@ import (
 	"context"
 	"errors"
 	"reflect"
+	"strconv"
 	"testing"
+	"time"
 
 	"github.com/livepeer/go-livepeer/ai/worker"
+	"github.com/livepeer/go-livepeer/core"
 )
 
 func Test_submitLLM(t *testing.T) {
@@ -125,6 +128,52 @@ func TestEncodeReqMetadata(t *testing.T) {
 	}
 }
 
+// Test_processAIRequest_NoOrchestrators_NonLive ensures that a non-live AI
+// request (liveParams == nil) returns a ServiceUnavailableError instead of
+// panicking when no orchestrators are available.
+func Test_processAIRequest_NoOrchestrators_NonLive(t *testing.T) {
+	cap := core.Capability_AudioToText
+	modelID := "openai/whisper-large-v3"
+
+	node := &core.LivepeerNode{
+		OrchestratorPool:          &stubDiscovery{},
+		AIProcesssingRetryTimeout: time.Second,
+	}
+
+	// Seed the selector cache with empty pools so Select returns nil without
+	// triggering a network refresh (empty OrchestratorPool keeps the refresh
+	// threshold at zero).
+	sel := &AISessionSelector{
+		warmPool:        NewAISessionPool(&LIFOSelector{}, newSuspender(), 0),
+		coldPool:        NewAISessionPool(&LIFOSelector{}, newSuspender(), 0),
+		ttl:             time.Hour,
+		lastRefreshTime: time.Now(),
+		cap:             cap,
+		modelID:         modelID,
+		node:            node,
+		suspender:       newSuspender(),
+	}
+	mgr := NewAISessionManager(node, time.Hour)
+	mgr.selectors[strconv.Itoa(int(cap))+"_"+modelID] = sel
+
+	params := aiRequestParams{
+		node:        node,
+		sessManager: mgr,
+		// liveParams intentionally nil to mimic a non-live pipeline request.
+	}
+
+	req := worker.GenAudioToTextMultipartRequestBody{ModelId: &modelID}
+
+	resp, err := processAIRequest(context.Background(), params, req)
+	if resp != nil {
+		t.Fatalf("expected nil response, got %v", resp)
+	}
+	var unavailable *ServiceUnavailableError
+	if !errors.As(err, &unavailable) {
+		t.Fatalf("expected ServiceUnavailableError, got %v", err)
+	}
+}
+
 func Test_isNoCapacityError(t *testing.T) {
 	tests := []struct {
 		name string