feat(streaming): configurable nginx, controller, and UI inactivity ti…

helm/kagent/files/nginx.conf

@@ -47,8 +47,8 @@ http {
             proxy_set_header X-Forwarded-Proto $scheme;
             proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
             proxy_set_header Origin $scheme://$host;
-            proxy_read_timeout 600s;
-            proxy_send_timeout 600s;
+            proxy_read_timeout {{ .Values.ui.nginx.proxyReadTimeout }};
+            proxy_send_timeout {{ .Values.ui.nginx.proxySendTimeout }};
             proxy_buffering off;
         }
 
@@ -63,6 +63,10 @@ http {
             proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
             proxy_set_header Origin $scheme://$host;
             proxy_cache_bypass $http_upgrade;
+            # Increased timeouts for streaming endpoints
+            proxy_read_timeout {{ .Values.ui.nginx.proxyReadTimeout }};
+            proxy_send_timeout {{ .Values.ui.nginx.proxySendTimeout }};
+            proxy_buffering off;
         }
 
         location /health {
@@ -80,8 +84,8 @@ http {
             proxy_set_header X-Forwarded-Proto $scheme;
             proxy_set_header X-Forwarded-Host $server_name;
             proxy_cache_bypass $http_upgrade;
-            proxy_read_timeout 3600s;
-            proxy_send_timeout 3600s;
+            proxy_read_timeout {{ .Values.ui.nginx.proxyReadTimeout }};
+            proxy_send_timeout {{ .Values.ui.nginx.proxySendTimeout }};
             proxy_buffering off;
         }
     }

helm/kagent/templates/openshift-route.yaml

@@ -6,6 +6,10 @@ kind: Route
 metadata:
     name: {{ include "kagent.fullname" . }}-ui
     namespace: {{ include "kagent.namespace" . }}
+    {{- with .Values.ui.openshiftRoute.annotations }}
+    annotations:
+      {{- toYaml . | nindent 6 }}
+    {{- end }}
 spec:
     to:
         kind: Service

helm/kagent/templates/ui-deployment.yaml

@@ -64,6 +64,10 @@ spec:
             - name: SSO_REDIRECT_PATH
               value: {{ .Values.ui.auth.ssoRedirectPath | default "/oauth2/start" | quote }}
             {{- end }}
+            {{- if .Values.ui.streamTimeoutSeconds }}
+            - name: KAGENT_STREAM_TIMEOUT_MS
+              value: {{ mul (int .Values.ui.streamTimeoutSeconds) 1000 | quote }}
+            {{- end }}
             {{- with .Values.ui.additionalForwardedHeaders }}
             - name: KAGENT_ADDITIONAL_FORWARDED_HEADERS
               value: {{ join "," . | quote }}

helm/kagent/values.yaml

@@ -328,6 +328,17 @@ ui:
     # Path to redirect users to when they click "Sign in with SSO" on the login page
     # Default: /oauth2/start (oauth2-proxy's authentication start endpoint)
     ssoRedirectPath: "/oauth2/start"
+  # -- Client-side chat stream inactivity timeout (seconds). The browser aborts a
+  # streaming response if no event is received within this window. Should be >=
+  # ui.nginx.proxyReadTimeout so nginx isn't the silent limit. Default 1800 (30m).
+  streamTimeoutSeconds: 1800
+  # -- Nginx proxy timeout configuration for the UI sidecar (values are passed
+  # directly to the corresponding nginx directives, e.g. "1800s").
+  nginx:
+    # -- proxy_read_timeout: max time between two successive reads from the upstream.
+    proxyReadTimeout: 1800s
+    # -- proxy_send_timeout: max time between two successive writes to the upstream.
+    proxySendTimeout: 1800s
   env: {} # Additional configuration key-value pairs for the ui ConfigMap
   # -- Additional request headers (beyond Authorization) the UI proxy will forward
   # to the backend. Names are case-insensitive. Hop-by-hop headers (Connection,
@@ -390,6 +401,13 @@ ui:
   #       path: /health
   #       port: http
   #     periodSeconds: 15
+
+  # OpenShift Route only (when route.openshift.io/v1 exists). Long timeouts are required
+  # for A2A/SSE streaming through the cluster router; defaults are often ~60s and cause
+  # net::ERR_INCOMPLETE_CHUNKED_ENCODING in the browser.
+  openshiftRoute:
+    annotations:
+      haproxy.router.openshift.io/timeout: 120m
 # ==============================================================================
 # LLM PROVIDERS CONFIGURATION
 # ==============================================================================

ui/src/app/a2a/[namespace]/[agentName]/route.ts

@@ -32,7 +32,7 @@ export async function POST(
 
     if (!backendResponse.ok) {
       const errorText = await backendResponse.text();
-      return new Response(errorText || 'Backend request failed', { 
+      return new Response(errorText || 'Backend request failed', {
         status: backendResponse.status,
         headers: {
           'Content-Type': 'text/plain',
@@ -90,6 +90,10 @@ export async function POST(
               return Promise.resolve();
             }
 
+            // Any upstream bytes count as activity for proxies; also start keep-alives
+            // before the first complete SSE frame (otherwise HAProxy may idle-timeout).
+            resetKeepAliveTimer();
+
             buffer += decoder.decode(value, { stream: true });
 
             // Process complete SSE events (delimited by \n\n)
@@ -111,7 +115,7 @@ export async function POST(
           }).catch(error => {
             console.error('A2A Proxy: Error in stream pump:', error);
             if (keepAliveTimer) clearTimeout(keepAliveTimer);
-            
+
             if (!isClosed) {
               controller.error(error);
               isClosed = true;
@@ -121,6 +125,9 @@ export async function POST(
           });
         };
 
+        // Begin keep-alives immediately so the browser↔UI connection survives long gaps
+        // until the first (or any) chunk from the controller.
+        resetKeepAliveTimer();
         pump();
       }
     });

ui/src/app/actions/config.ts

@@ -0,0 +1,19 @@
+"use server";
+
+import { DEFAULT_STREAM_TIMEOUT_MS } from "@/lib/constants";
+
+export interface UiRuntimeConfig {
+  streamTimeoutMs: number;
+}
+
+/**
+ * Returns runtime UI configuration sourced from server-side environment
+ * variables (set by the Helm chart). Read on the server so values reflect the
+ * deployment at runtime, unlike NEXT_PUBLIC_* vars which are inlined at build.
+ */
+export async function getUiRuntimeConfig(): Promise<UiRuntimeConfig> {
+  const raw = process.env.KAGENT_STREAM_TIMEOUT_MS;
+  const parsed = raw ? Number(raw) : NaN;
+  const streamTimeoutMs = Number.isFinite(parsed) && parsed > 0 ? parsed : DEFAULT_STREAM_TIMEOUT_MS;
+  return { streamTimeoutMs };
+}

ui/src/components/chat/ChatInterface.tsx

@@ -20,6 +20,8 @@ import type { TokenStats, Session, ChatStatus, ToolDecision } from "@/types";
 import StatusDisplay from "./StatusDisplay";
 import { createSession, getSessionTasks, checkSessionExists } from "@/app/actions/sessions";
 import { waitForSandboxAgentReady } from "@/app/actions/agents";
+import { getUiRuntimeConfig } from "@/app/actions/config";
+import { DEFAULT_STREAM_TIMEOUT_MS } from "@/lib/constants";
 import { toast } from "sonner";
 import { useRouter } from "next/navigation";
 import { createMessageHandlers, extractMessagesFromTasks, extractApprovalMessagesFromTasks, extractTokenStatsFromTasks, createMessage, ADKMetadata, ProcessedToolCallData } from "@/lib/messageHandlers";
@@ -65,6 +67,22 @@ export default function ChatInterface({ selectedAgentName, selectedNamespace, se
   const pendingDecisionsRef = useRef<Record<string, ToolDecision>>({});
   /** Per-tool rejection reasons collected as the user rejects individual tools. */
   const pendingRejectionReasonsRef = useRef<Record<string, string>>({});
+  // Stream inactivity timeout (ms), configurable via Helm (ui.streamTimeoutSeconds).
+  const streamTimeoutMsRef = useRef<number>(DEFAULT_STREAM_TIMEOUT_MS);
+
+  useEffect(() => {
+    let cancelled = false;
+    getUiRuntimeConfig()
+      .then((config) => {
+        if (!cancelled) streamTimeoutMsRef.current = config.streamTimeoutMs;
+      })
+      .catch(() => {
+        /* keep default on failure */
+      });
+    return () => {
+      cancelled = true;
+    };
+  }, []);
 
   const {
     isListening,
@@ -314,18 +332,24 @@ export default function ChatInterface({ selectedAgentName, selectedNamespace, se
   const consumeStream = async (stream: AsyncIterable<unknown>) => {
     let timeoutTimer: NodeJS.Timeout | null = null;
     let streamActive = true;
-    const STREAM_TIMEOUT_MS = 600000; // 10 minutes
+
+    const formatTimeout = (ms: number): string => {
+      const mins = ms / 60000;
+      return mins >= 1 ? `${Math.ceil(mins)} minutes` : `${Math.round(ms / 1000)} seconds`;
+    };
 
     const startTimeout = () => {
       if (timeoutTimer) clearTimeout(timeoutTimer);
+      const streamTimeoutMs = streamTimeoutMsRef.current;
       timeoutTimer = setTimeout(() => {
         if (streamActive) {
-          console.error("⏰ Stream timeout - no events received for 10 minutes");
-          toast.error("⏰ Stream timed out - no events received for 10 minutes");
+          const label = formatTimeout(streamTimeoutMs);
+          console.error(`⏰ Stream timeout - no events received for ${label}`);
+          toast.error(`⏰ Stream timed out - no events received for ${label}`);
           streamActive = false;
           abortControllerRef.current?.abort();
         }
-      }, STREAM_TIMEOUT_MS);
+      }, streamTimeoutMs);
     };
     startTimeout();
 

ui/src/lib/constants.ts

@@ -1,3 +1,7 @@
 // Model-related constants
 export const OLLAMA_DEFAULT_TAG = "latest";
 export const OLLAMA_DEFAULT_HOST = "localhost:11434";
+
+// Default client-side stream inactivity timeout (30 minutes) used when Helm
+// does not provide an override via ui.streamTimeoutSeconds.
+export const DEFAULT_STREAM_TIMEOUT_MS = 1800000;