fix(backend): fix kubeconfig injection for slow-starting workspaces (CRW-11193) (#1620)

* fix(backend): start polling fallback when watch times out in PostStartInjector

The 60-second watch timeout called cleanup() and silently abandoned
injection without starting the polling fallback. Any workspace that takes
more than 60 seconds to reach Running — which is common — never received
kubeconfig or podman credentials.

Fix: start startPollingFallback() from the timeout handler so injection
can still complete via GET polling (10 s interval, up to 5 minutes).

Fixes: https://redhat.atlassian.net/browse/CRW-11193

Assisted-by: Claude Sonnet 4.6
Signed-off-by: Oleksii Orel <oorel@redhat.com>

* fix(backend): increase watch timeout to 120s and log polling elapsed time

- Raise INJECTION_TIMEOUT_MS from 60s to 120s. The watch is a cheap
  long-lived HTTP connection, while the polling fallback fires a GET
  every 10 seconds. Keeping the watch alive longer reduces unnecessary
  polling for workspaces that start between 60s and 120s.

- Log how long the workspace was in Starting phase when injection
  completes via polling (e.g. "workspace ready after 95s"), which helps
  operators diagnose slow-start clusters.

- Add a comment in startPollingFallback() explaining why the same
  devworkspaceApi instance is safe to reuse across watch and polling
  (getByName is a standalone REST call, independent of watch state).

Assists: https://redhat.atlassian.net/browse/CRW-11193

Assisted-by: Claude Sonnet 4.6
Signed-off-by: Oleksii Orel <oorel@redhat.com>

* fix(backend): increase watch timeout to 300s to match startTimeout

Align INJECTION_TIMEOUT_MS with the server-side startTimeout default
(300 s). The watch is a cheap long-lived HTTP connection that should
stay alive for the full startup window rather than falling back to
polling at 120 s.

Assists: https://redhat.atlassian.net/browse/CRW-11193

Assisted-by: Claude Sonnet 4.6
Signed-off-by: Oleksii Orel <oorel@redhat.com>

---------

Signed-off-by: Oleksii Orel <oorel@redhat.com>

Author: olexii4

packages/dashboard-backend/src/services/PostStartInjector.ts

@@ -16,7 +16,7 @@ import { IDevWorkspaceApi, IKubeConfigApi, IPodmanApi } from '@/devworkspaceClie
 import { MessageListener } from '@/services/types/Observer';
 import { logger } from '@/utils/logger';
 
-const INJECTION_TIMEOUT_MS = 60000;
+const INJECTION_TIMEOUT_MS = 300000;
 const POLL_INTERVAL_MS = 10000;
 const POLL_TIMEOUT_MS = 300000;
 
@@ -68,8 +68,15 @@ export class PostStartInjector {
     PostStartInjector.activeWatches.set(key, cleanup);
 
     const timeoutHandle = setTimeout(() => {
-      logger.warn(`PostStartInjector: timed out waiting for ${key} to reach Running`);
+      logger.warn(`PostStartInjector: watch timed out for ${key}, starting polling fallback`);
       cleanup();
+      PostStartInjector.startPollingFallback(
+        namespace,
+        workspaceName,
+        devworkspaceApi,
+        kubeConfigApi,
+        podmanApi,
+      );
     }, INJECTION_TIMEOUT_MS);
 
     const cleanupWithTimeout = () => {
@@ -157,6 +164,9 @@ export class PostStartInjector {
     };
 
     // Re-register so duplicate watchAndInject calls are still blocked during polling.
+    // Note: the same devworkspaceApi instance is reused here. stopWatching() was already
+    // called before startPollingFallback(), but getByName() is a standalone REST call
+    // with no dependency on watch state, so the instance is safe to reuse.
     PostStartInjector.activeWatches.set(key, cleanup);
 
     let elapsed = 0;
@@ -195,6 +205,7 @@ export class PostStartInjector {
               kubeConfigApi,
               podmanApi,
               key,
+              elapsed,
             );
           } else {
             logger.info(`PostStartInjector: ${key} is in phase ${phase}, stopping poll`);
@@ -212,8 +223,13 @@ export class PostStartInjector {
     kubeConfigApi: IKubeConfigApi,
     podmanApi: IPodmanApi,
     key: string,
+    elapsedMs?: number,
   ): Promise<void> {
-    logger.info(`PostStartInjector: ${key} is Running, injecting kubeconfig and podman login`);
+    const elapsed =
+      elapsedMs !== undefined ? ` (workspace ready after ${Math.round(elapsedMs / 1000)}s)` : '';
+    logger.info(
+      `PostStartInjector: ${key} is Running, injecting kubeconfig and podman login${elapsed}`,
+    );
     try {
       await kubeConfigApi.injectKubeConfig(namespace, devworkspaceId);
     } catch (e) {

packages/dashboard-backend/src/services/__tests__/PostStartInjector.spec.ts

@@ -196,12 +196,34 @@ describe('PostStartInjector', () => {
 
   // ── timeout ───────────────────────────────────────────────────────────────
 
-  test('cleans up on 60 s timeout', () => {
+  test('stops watch and starts polling fallback on 300 s timeout', () => {
+    (devworkspaceApi.getByName as jest.Mock).mockResolvedValue({
+      status: { phase: 'Starting' },
+    });
+
     invoke();
-    jest.advanceTimersByTime(60000);
+    jest.advanceTimersByTime(300000);
 
+    // Watch stopped
     expect(devworkspaceApi.stopWatching).toHaveBeenCalled();
-    expect((PostStartInjector as any).activeWatches.has(key)).toBe(false);
+    // Polling fallback re-registers the key so injection can still complete
+    expect((PostStartInjector as any).activeWatches.has(key)).toBe(true);
+  });
+
+  test('injects credentials via polling fallback after 60 s watch timeout', async () => {
+    (devworkspaceApi.getByName as jest.Mock).mockResolvedValue({
+      status: { phase: 'Running', devworkspaceId: 'ws-timeout-id' },
+    });
+
+    invoke();
+    jest.advanceTimersByTime(300000);
+
+    // Advance polling interval so getByName is called
+    jest.advanceTimersByTime(10000);
+    await Promise.resolve().then(() => Promise.resolve());
+
+    expect(kubeConfigApi.injectKubeConfig).toHaveBeenCalledWith(namespace, 'ws-timeout-id');
+    expect(podmanApi.podmanLogin).toHaveBeenCalledWith(namespace, 'ws-timeout-id');
   });
 
   // ── polling fallback (triggered by ERROR event) ───────────────────────────