Add out-of-memory session recovery (#491)

BunsDev · web-flow · commit 13c39430c3ad · 2026-04-21T22:11:09.000-05:00
* fix: verify published cli with okcode bin

* fix: make npm publish rerun-safe

* fix: publish cli from cjs bin

* Add out-of-memory session recovery
diff --git a/apps/web/src/components/ChatView.browser.tsx b/apps/web/src/components/ChatView.browser.tsx
@@ -291,6 +291,34 @@ function buildFixture(snapshot: OrchestrationReadModel): TestFixture {
   };
 }
 
+function withThreadSessionError(
+  snapshot: OrchestrationReadModel,
+  input: {
+    status: OrchestrationSessionStatus;
+    lastError: string;
+  },
+): OrchestrationReadModel {
+  return {
+    ...snapshot,
+    threads: snapshot.threads.map((thread) =>
+      thread.id === THREAD_ID && thread.session
+        ? {
+            ...thread,
+            session: {
+              threadId: thread.session.threadId,
+              providerName: thread.session.providerName,
+              runtimeMode: thread.session.runtimeMode,
+              activeTurnId: thread.session.activeTurnId,
+              status: input.status,
+              lastError: input.lastError,
+              updatedAt: NOW_ISO,
+            },
+          }
+        : thread,
+    ),
+  };
+}
+
 function addThreadToSnapshot(
   snapshot: OrchestrationReadModel,
   threadId: ThreadId,
@@ -1634,6 +1662,50 @@ describe("ChatView timeline estimator parity (full app)", () => {
     }
   });
 
+  it("resets the provider session from the error banner after an out-of-memory failure", async () => {
+    wsRequests.length = 0;
+
+    const mounted = await mountChatView({
+      viewport: DEFAULT_VIEWPORT,
+      snapshot: withThreadSessionError(
+        createSnapshotForTargetUser({
+          targetMessageId: "msg-user-oom-reset" as MessageId,
+          targetText: "oom reset target",
+        }),
+        {
+          status: "error",
+          lastError:
+            "FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory",
+        },
+      ),
+    });
+
+    try {
+      const recoverButton = await waitForElement(
+        () =>
+          document.querySelector<HTMLButtonElement>(
+            'button[aria-label="Reset session after out-of-memory failure"]',
+          ),
+        "Unable to find out-of-memory recovery button.",
+      );
+
+      recoverButton.click();
+
+      await vi.waitFor(
+        () =>
+          wsRequests.some(
+            (request) =>
+              request._tag === ORCHESTRATION_WS_METHODS.dispatchCommand &&
+              request.type === "thread.session.stop" &&
+              request.threadId === THREAD_ID,
+          ),
+        { timeout: 8_000, interval: 16 },
+      );
+    } finally {
+      await mounted.cleanup();
+    }
+  });
+
   it("keeps the new thread selected after clicking the new-thread button", async () => {
     const mounted = await mountChatView({
       viewport: DEFAULT_VIEWPORT,
diff --git a/apps/web/src/components/ChatView.tsx b/apps/web/src/components/ChatView.tsx
@@ -3880,6 +3880,17 @@ export default function ChatView({
     });
   };
 
+  const onRecoverFromOutOfMemory = async () => {
+    const api = readNativeApi();
+    if (!api || !activeThread || isRemoteActionBlocked) return;
+    await api.orchestration.dispatchCommand({
+      type: "thread.session.stop",
+      commandId: newCommandId(),
+      threadId: activeThread.id,
+      createdAt: new Date().toISOString(),
+    });
+  };
+
   const onClearQueue = useCallback(() => {
     setOptimisticUserMessages((existing) => {
       for (const msg of existing) {
@@ -4955,6 +4966,7 @@ export default function ChatView({
         showNotificationDetails={settings.showNotificationDetails}
         includeDiagnosticsTipsInCopy={settings.includeDiagnosticsTipsInCopy}
         onDismissThreadError={() => setThreadError(activeThread.id, null)}
+        onRecoverFromOutOfMemory={() => void onRecoverFromOutOfMemory()}
         providerStatus={activeProviderStatus}
         transportState={transportState}
         isMobileCompanion={isMobileCompanion}
diff --git a/apps/web/src/components/chat/ErrorNotificationBar.test.tsx b/apps/web/src/components/chat/ErrorNotificationBar.test.tsx
@@ -24,7 +24,8 @@ const THREAD_ERROR =
 function renderBar(
   overrides: Partial<ComponentProps<typeof ErrorNotificationBar>> = {},
 ): ReactElement {
-  const { onDismissThreadError, transportState, ...restOverrides } = overrides;
+  const { onDismissThreadError, onRecoverFromOutOfMemory, transportState, ...restOverrides } =
+    overrides;
   return (
     <ErrorNotificationBar
       threadError={THREAD_ERROR}
@@ -35,6 +36,7 @@ function renderBar(
       isMobileCompanion={false}
       {...restOverrides}
       {...(onDismissThreadError ? { onDismissThreadError } : {})}
+      {...(onRecoverFromOutOfMemory ? { onRecoverFromOutOfMemory } : {})}
       {...(transportState ? { transportState } : {})}
     />
   );
@@ -86,72 +88,28 @@ describe("ErrorNotificationBar", () => {
     expect(markup).toContain("Base branch &#x27;main&#x27; does not resolve to a commit yet.");
   });
 
-  it("re-shows thread errors when the message changes after dismissal", async () => {
-    const onDismissThreadError = vi.fn();
+  it("shows an out-of-memory recovery action when the thread error is recoverable", async () => {
+    const onRecoverFromOutOfMemory = vi.fn();
     let renderer: ReactTestRenderer | null = null;
-
     await act(async () => {
       renderer = create(
-        <ErrorNotificationBar
-          threadError={THREAD_ERROR}
-          showAuthFailuresAsErrors
-          showNotificationDetails={false}
-          includeDiagnosticsTipsInCopy={false}
-          providerStatus={null}
-          isMobileCompanion={false}
-          onDismissThreadError={onDismissThreadError}
-        />,
-      );
-    });
-
-    const dismissAll = renderer!.root.findByProps({ "aria-label": "Dismiss notifications" });
-    await act(async () => {
-      dismissAll.props.onClick();
-    });
-
-    expect(renderer!.toJSON()).toBeNull();
-
-    await act(async () => {
-      renderer!.update(
-        <ErrorNotificationBar
-          threadError="Codex CLI is not authenticated. Run `codex login` and try again."
-          showAuthFailuresAsErrors
-          showNotificationDetails={false}
-          includeDiagnosticsTipsInCopy={false}
-          providerStatus={null}
-          isMobileCompanion={false}
-          onDismissThreadError={onDismissThreadError}
-        />,
+        renderBar({
+          threadError:
+            "FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory",
+          onRecoverFromOutOfMemory,
+        }),
       );
     });
 
-    expect(renderer!.toJSON()).not.toBeNull();
-    expect(renderer!.root.findByProps({ "aria-label": "Show 1 notification" })).toBeTruthy();
-  });
-
-  it("does not hide non-dismissible provider notifications via dismiss all", async () => {
-    let renderer: ReactTestRenderer | null = null;
-
-    await act(async () => {
-      renderer = create(
-        <ErrorNotificationBar
-          threadError={null}
-          showAuthFailuresAsErrors
-          showNotificationDetails={false}
-          includeDiagnosticsTipsInCopy={false}
-          providerStatus={makeProviderStatus()}
-          isMobileCompanion={false}
-        />,
-      );
+    const root = renderer!.root;
+    const recoverButton = root.findByProps({
+      "aria-label": "Reset session after out-of-memory failure",
     });
 
-    const dismissAll = renderer!.root.findByProps({ "aria-label": "Dismiss notifications" });
     await act(async () => {
-      dismissAll.props.onClick();
+      recoverButton.props.onClick();
     });
 
-    expect(renderer!.toJSON()).not.toBeNull();
-    expect(renderer!.root.findByProps({ "aria-label": "Show 1 notification" })).toBeTruthy();
-    expect(JSON.stringify(renderer!.toJSON())).toContain("OpenAI (Codex CLI) needs verification");
+    expect(onRecoverFromOutOfMemory).toHaveBeenCalledTimes(1);
   });
 });
diff --git a/apps/web/src/components/chat/ErrorNotificationBar.tsx b/apps/web/src/components/chat/ErrorNotificationBar.tsx
@@ -13,6 +13,7 @@ import {
   buildThreadErrorDiagnosticsCopy,
   humanizeThreadError,
   isAuthenticationThreadError,
+  isOutOfMemoryThreadError,
 } from "./threadError";
 import {
   getProviderStatusHeading,
@@ -33,6 +34,8 @@ interface ErrorNotificationBarProps {
   includeDiagnosticsTipsInCopy?: boolean;
   /** Dismiss the thread error */
   onDismissThreadError?: () => void;
+  /** Reset a provider session after an OOM failure */
+  onRecoverFromOutOfMemory?: () => void;
   /** Provider health status */
   providerStatus: ServerProviderStatus | null;
   /** Companion transport state (only relevant for mobile companion) */
@@ -49,6 +52,9 @@ interface NotificationItem {
   description: string;
   detailsText?: string | null;
   diagnosticsCopyText?: string | null;
+  actionLabel?: string;
+  actionAriaLabel?: string;
+  onAction?: () => void;
   severity: "error" | "warning" | "info";
   dismissible: boolean;
   onDismiss?: () => void;
@@ -64,6 +70,7 @@ export const ErrorNotificationBar = memo(function ErrorNotificationBar({
   showNotificationDetails = false,
   includeDiagnosticsTipsInCopy = false,
   onDismissThreadError,
+  onRecoverFromOutOfMemory,
   providerStatus,
   transportState,
   isMobileCompanion,
@@ -133,6 +140,8 @@ export const ErrorNotificationBar = memo(function ErrorNotificationBar({
     if (threadError) {
       if (showAuthFailuresAsErrors || !isAuthenticationThreadError(threadError)) {
         const presentation = humanizeThreadError(threadError);
+        const showOutOfMemoryRecovery =
+          isOutOfMemoryThreadError(threadError) && onRecoverFromOutOfMemory !== undefined;
         items.push({
           id: buildThreadErrorNotificationId(threadError),
           kind: "thread-error",
@@ -143,6 +152,13 @@ export const ErrorNotificationBar = memo(function ErrorNotificationBar({
           diagnosticsCopyText: buildThreadErrorDiagnosticsCopy(threadError, {
             includeTips: includeDiagnosticsTipsInCopy,
           }),
+          ...(showOutOfMemoryRecovery
+            ? {
+                actionLabel: "Reset session",
+                actionAriaLabel: "Reset session after out-of-memory failure",
+                onAction: onRecoverFromOutOfMemory,
+              }
+            : {}),
           severity: "error",
           dismissible: !!onDismissThreadError,
           ...(onDismissThreadError ? { onDismiss: onDismissThreadError } : {}),
@@ -156,6 +172,7 @@ export const ErrorNotificationBar = memo(function ErrorNotificationBar({
     showAuthFailuresAsErrors,
     includeDiagnosticsTipsInCopy,
     onDismissThreadError,
+    onRecoverFromOutOfMemory,
     providerStatus,
     transportState,
     isMobileCompanion,
@@ -221,6 +238,9 @@ export const ErrorNotificationBar = memo(function ErrorNotificationBar({
   if (visibleNotifications.length === 0) return null;
 
   const primary = visibleNotifications[0]!;
+  const actionNotification = visibleNotifications.find(
+    (notification) => notification.onAction && notification.actionLabel,
+  );
   const PrimaryIcon = primary.icon;
   const count = visibleNotifications.length;
   const countLabel = count === 1 ? "1 notification" : `${count} notifications`;
@@ -261,6 +281,18 @@ export const ErrorNotificationBar = memo(function ErrorNotificationBar({
           </span>
 
           <div className="flex shrink-0 items-center gap-1">
+            {actionNotification?.onAction && actionNotification.actionLabel ? (
+              <Button
+                type="button"
+                variant="outline"
+                size="xs"
+                aria-label={actionNotification.actionAriaLabel ?? actionNotification.actionLabel}
+                className="min-w-0 px-2 text-[10px] font-medium"
+                onClick={actionNotification.onAction}
+              >
+                {actionNotification.actionLabel}
+              </Button>
+            ) : null}
             <Button
               type="button"
               variant="outline"
@@ -313,6 +345,18 @@ export const ErrorNotificationBar = memo(function ErrorNotificationBar({
                     ) : null}
                   </div>
                   <div className="mt-0.5 flex shrink-0 items-center gap-1">
+                    {notif.onAction && notif.actionLabel ? (
+                      <Button
+                        type="button"
+                        variant="outline"
+                        size="xs"
+                        aria-label={notif.actionAriaLabel ?? notif.actionLabel}
+                        className="h-6 px-2 text-[10px]"
+                        onClick={notif.onAction}
+                      >
+                        {notif.actionLabel}
+                      </Button>
+                    ) : null}
                     {notif.kind === "thread-error" && notif.diagnosticsCopyText ? (
                       <MessageCopyButton
                         text={notif.diagnosticsCopyText}
diff --git a/apps/web/src/components/chat/threadError.test.ts b/apps/web/src/components/chat/threadError.test.ts
@@ -3,6 +3,7 @@ import { describe, expect, it } from "vitest";
 import {
   buildThreadErrorDiagnosticsCopy,
   humanizeThreadError,
+  isOutOfMemoryThreadError,
   isAuthenticationThreadError,
 } from "./threadError";
 
@@ -57,6 +58,18 @@ describe("humanizeThreadError", () => {
     expect(isAuthenticationThreadError("Provider crashed while starting.")).toBe(false);
   });
 
+  it("detects out-of-memory failures", () => {
+    expect(
+      isOutOfMemoryThreadError(
+        "Provider crashed: FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory",
+      ),
+    ).toBe(true);
+    expect(
+      isOutOfMemoryThreadError("Process exited: memory limit exceeded while streaming turn"),
+    ).toBe(true);
+    expect(isOutOfMemoryThreadError("Provider crashed while starting.")).toBe(false);
+  });
+
   it("builds redacted diagnostics copy without optional tips by default", () => {
     expect(
       buildThreadErrorDiagnosticsCopy(
diff --git a/apps/web/src/components/chat/threadError.ts b/apps/web/src/components/chat/threadError.ts
