fix(node-agent): resilient output upload — 60s HTTP timeout + 4× retry on upload-URL

After a CPU-heavy AE render+transcode the orchestrator/DB can be briefly slow;
the 15s client timeout made the post-render output-upload-url call fail and the
finished MP4 was dropped (completed without export). Bumped client timeout to 60s
and retry the upload-URL call up to 4× with backoff so a finished render's output
is never lost to a transient stall.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

services/node-agent/cmd/agent/main.go

@@ -472,9 +472,23 @@ func (a *Agent) runJob(ctx context.Context, job *client.ClaimedJob) {
 	uploadCtx, uploadCancel := context.WithTimeout(context.Background(), 5*time.Minute)
 	defer uploadCancel()
 
-	uploadInfo, urlErr := a.orch.GetOutputUploadURL(uploadCtx, job.JobID)
+	// Retry the upload-URL call: right after a CPU-heavy render the orchestrator/DB
+	// can be briefly slow, and dropping a finished render's output is the worst outcome.
+	var uploadInfo *client.OutputUploadURLResponse
+	var urlErr error
+	for attempt := 1; attempt <= 4; attempt++ {
+		uploadInfo, urlErr = a.orch.GetOutputUploadURL(uploadCtx, job.JobID)
+		if urlErr == nil {
+			break
+		}
+		log.Printf("[job %s] get upload URL attempt %d failed: %v", job.JobID, attempt, urlErr)
+		select {
+		case <-uploadCtx.Done():
+		case <-time.After(time.Duration(attempt*3) * time.Second):
+		}
+	}
 	if urlErr != nil {
-		log.Printf("[job %s] get upload URL failed: %v — completing without export", job.JobID, urlErr)
+		log.Printf("[job %s] get upload URL failed after retries: %v — completing without export", job.JobID, urlErr)
 	} else {
 		log.Printf("[job %s] uploading output to %s", job.JobID, uploadInfo.ObjectKey)
 		if _, upErr := runner.UploadFile(uploadCtx, uploadInfo.UploadURL, outputPath); upErr != nil {

services/node-agent/internal/client/client.go

@@ -26,7 +26,9 @@ func New(baseURL, nodeHMACSecret string) *Client {
 	return &Client{
 		base:   strings.TrimRight(baseURL, "/"),
 		secret: nodeHMACSecret,
-		http:   &http.Client{Timeout: 15 * time.Second},
+		// 60s: the post-render output-upload-url call (export insert + presign) can be
+		// slow when the DB is briefly busy right after a CPU-heavy render/transcode.
+		http:   &http.Client{Timeout: 60 * time.Second},
 	}
 }