fix(core): add retry logic for transient SSL/TLS errors (#17318) (#18310)

packages/core/src/core/geminiChat.ts

@@ -394,16 +394,23 @@ export class GeminiChat {
               return; // Stop the generator
             }
 
-            if (isConnectionPhase) {
+            // Check if the error is retryable (e.g., transient SSL errors
-              throw error;
+            // like ERR_SSL_SSLV3_ALERT_BAD_RECORD_MAC)
-            }
-            lastError = error;
-            const isContentError = error instanceof InvalidStreamError;
             const isRetryable = isRetryableError(
               error,
               this.config.getRetryFetchErrors(),
             );
 
+            // For connection phase errors, only retryable errors should continue
+            if (isConnectionPhase) {
+              if (!isRetryable || signal.aborted) {
+                throw error;
+              }
+              // Fall through to retry logic for retryable connection errors
+            }
+            lastError = error;
+            const isContentError = error instanceof InvalidStreamError;
+
             if (
               (isContentError && isGemini2Model(model)) ||
               (isRetryable && !signal.aborted)

packages/core/src/core/geminiChat_network_retry.test.ts

@@ -274,4 +274,204 @@ describe('GeminiChat Network Retries', () => {
 
     expect(mockLogContentRetry).not.toHaveBeenCalled();
   });
+
+  it('should retry on SSL error during connection phase (ERR_SSL_SSLV3_ALERT_BAD_RECORD_MAC)', async () => {
+    // Create an SSL error that occurs during connection (before any yield)
+    const sslError = new Error(
+      'SSL routines:ssl3_read_bytes:sslv3 alert bad record mac',
+    );
+    (sslError as NodeJS.ErrnoException).code =
+      'ERR_SSL_SSLV3_ALERT_BAD_RECORD_MAC';
+
+    vi.mocked(mockContentGenerator.generateContentStream)
+      // First call: throw SSL error immediately (connection phase)
+      .mockRejectedValueOnce(sslError)
+      // Second call: succeed
+      .mockImplementationOnce(async () =>
+        (async function* () {
+          yield {
+            candidates: [
+              {
+                content: { parts: [{ text: 'Success after SSL retry' }] },
+                finishReason: 'STOP',
+              },
+            ],
+          } as unknown as GenerateContentResponse;
+        })(),
+      );
+
+    const stream = await chat.sendMessageStream(
+      { model: 'test-model' },
+      'test message',
+      'prompt-id-ssl-retry',
+      new AbortController().signal,
+    );
+
+    const events: StreamEvent[] = [];
+    for await (const event of stream) {
+      events.push(event);
+    }
+
+    // Should have retried and succeeded
+    const retryEvent = events.find((e) => e.type === StreamEventType.RETRY);
+    expect(retryEvent).toBeDefined();
+
+    const successChunk = events.find(
+      (e) =>
+        e.type === StreamEventType.CHUNK &&
+        e.value.candidates?.[0]?.content?.parts?.[0]?.text ===
+          'Success after SSL retry',
+    );
+    expect(successChunk).toBeDefined();
+
+    // Verify the API was called twice (initial + retry)
+    expect(mockContentGenerator.generateContentStream).toHaveBeenCalledTimes(2);
+  });
+
+  it('should retry on ECONNRESET error during connection phase', async () => {
+    const connectionError = new Error('read ECONNRESET');
+    (connectionError as NodeJS.ErrnoException).code = 'ECONNRESET';
+
+    vi.mocked(mockContentGenerator.generateContentStream)
+      .mockRejectedValueOnce(connectionError)
+      .mockImplementationOnce(async () =>
+        (async function* () {
+          yield {
+            candidates: [
+              {
+                content: {
+                  parts: [{ text: 'Success after connection retry' }],
+                },
+                finishReason: 'STOP',
+              },
+            ],
+          } as unknown as GenerateContentResponse;
+        })(),
+      );
+
+    const stream = await chat.sendMessageStream(
+      { model: 'test-model' },
+      'test message',
+      'prompt-id-connection-retry',
+      new AbortController().signal,
+    );
+
+    const events: StreamEvent[] = [];
+    for await (const event of stream) {
+      events.push(event);
+    }
+
+    const retryEvent = events.find((e) => e.type === StreamEventType.RETRY);
+    expect(retryEvent).toBeDefined();
+
+    const successChunk = events.find(
+      (e) =>
+        e.type === StreamEventType.CHUNK &&
+        e.value.candidates?.[0]?.content?.parts?.[0]?.text ===
+          'Success after connection retry',
+    );
+    expect(successChunk).toBeDefined();
+  });
+
+  it('should NOT retry on non-retryable error during connection phase', async () => {
+    const nonRetryableError = new Error('Some non-retryable error');
+
+    vi.mocked(mockContentGenerator.generateContentStream).mockRejectedValueOnce(
+      nonRetryableError,
+    );
+
+    const stream = await chat.sendMessageStream(
+      { model: 'test-model' },
+      'test message',
+      'prompt-id-no-connection-retry',
+      new AbortController().signal,
+    );
+
+    await expect(async () => {
+      for await (const _ of stream) {
+        // consume
+      }
+    }).rejects.toThrow(nonRetryableError);
+
+    // Should only be called once (no retry)
+    expect(mockContentGenerator.generateContentStream).toHaveBeenCalledTimes(1);
+  });
+
+  it('should retry on SSL error during stream iteration (mid-stream failure)', async () => {
+    // This simulates the exact scenario from issue #17318 where the error
+    // occurs during a long session while streaming content
+    const sslError = new Error(
+      'request to https://cloudcode-pa.googleapis.com/v1internal:streamGenerateContent failed',
+    ) as NodeJS.ErrnoException & { type?: string };
+    sslError.type = 'system';
+    sslError.errno = 'ERR_SSL_SSLV3_ALERT_BAD_RECORD_MAC' as unknown as number;
+    sslError.code = 'ERR_SSL_SSLV3_ALERT_BAD_RECORD_MAC';
+
+    vi.mocked(mockContentGenerator.generateContentStream)
+      // First call: yield some content, then throw SSL error mid-stream
+      .mockImplementationOnce(async () =>
+        (async function* () {
+          yield {
+            candidates: [
+              { content: { parts: [{ text: 'Partial response...' }] } },
+            ],
+          } as unknown as GenerateContentResponse;
+          // SSL error occurs while waiting for more data
+          throw sslError;
+        })(),
+      )
+      // Second call: succeed
+      .mockImplementationOnce(async () =>
+        (async function* () {
+          yield {
+            candidates: [
+              {
+                content: { parts: [{ text: 'Complete response after retry' }] },
+                finishReason: 'STOP',
+              },
+            ],
+          } as unknown as GenerateContentResponse;
+        })(),
+      );
+
+    const stream = await chat.sendMessageStream(
+      { model: 'test-model' },
+      'test message',
+      'prompt-id-ssl-mid-stream',
+      new AbortController().signal,
+    );
+
+    const events: StreamEvent[] = [];
+    for await (const event of stream) {
+      events.push(event);
+    }
+
+    // Should have received partial content, then retry, then success
+    const partialChunk = events.find(
+      (e) =>
+        e.type === StreamEventType.CHUNK &&
+        e.value.candidates?.[0]?.content?.parts?.[0]?.text ===
+          'Partial response...',
+    );
+    expect(partialChunk).toBeDefined();
+
+    const retryEvent = events.find((e) => e.type === StreamEventType.RETRY);
+    expect(retryEvent).toBeDefined();
+
+    const successChunk = events.find(
+      (e) =>
+        e.type === StreamEventType.CHUNK &&
+        e.value.candidates?.[0]?.content?.parts?.[0]?.text ===
+          'Complete response after retry',
+    );
+    expect(successChunk).toBeDefined();
+
+    // Verify retry logging was called with NETWORK_ERROR type
+    expect(mockLogContentRetry).toHaveBeenCalledWith(
+      expect.anything(),
+      expect.objectContaining({
+        error_type: 'NETWORK_ERROR',
+      }),
+    );
+  });
 });

packages/core/src/utils/retry.test.ts

@@ -409,6 +409,87 @@ describe('retryWithBackoff', () => {
       await vi.runAllTimersAsync();
       await expect(promise).resolves.toBe('success');
     });
+
+    it('should retry on SSL error code (ERR_SSL_SSLV3_ALERT_BAD_RECORD_MAC)', async () => {
+      const error = new Error('SSL error');
+      (error as any).code = 'ERR_SSL_SSLV3_ALERT_BAD_RECORD_MAC';
+      const mockFn = vi
+        .fn()
+        .mockRejectedValueOnce(error)
+        .mockResolvedValue('success');
+
+      const promise = retryWithBackoff(mockFn, {
+        initialDelayMs: 1,
+        maxDelayMs: 1,
+      });
+      await vi.runAllTimersAsync();
+      await expect(promise).resolves.toBe('success');
+      expect(mockFn).toHaveBeenCalledTimes(2);
+    });
+
+    it('should retry on SSL error code in deeply nested cause chain', async () => {
+      const deepCause = new Error('OpenSSL error');
+      (deepCause as any).code = 'ERR_SSL_BAD_RECORD_MAC';
+
+      const middleCause = new Error('TLS handshake failed');
+      (middleCause as any).cause = deepCause;
+
+      const outerError = new Error('fetch failed');
+      (outerError as any).cause = middleCause;
+
+      const mockFn = vi
+        .fn()
+        .mockRejectedValueOnce(outerError)
+        .mockResolvedValue('success');
+
+      const promise = retryWithBackoff(mockFn, {
+        initialDelayMs: 1,
+        maxDelayMs: 1,
+      });
+      await vi.runAllTimersAsync();
+      await expect(promise).resolves.toBe('success');
+      expect(mockFn).toHaveBeenCalledTimes(2);
+    });
+
+    it('should retry on EPROTO error (generic protocol/SSL error)', async () => {
+      const error = new Error('Protocol error');
+      (error as any).code = 'EPROTO';
+      const mockFn = vi
+        .fn()
+        .mockRejectedValueOnce(error)
+        .mockResolvedValue('success');
+
+      const promise = retryWithBackoff(mockFn, {
+        initialDelayMs: 1,
+        maxDelayMs: 1,
+      });
+      await vi.runAllTimersAsync();
+      await expect(promise).resolves.toBe('success');
+      expect(mockFn).toHaveBeenCalledTimes(2);
+    });
+
+    it('should retry on gaxios-style SSL error with code property', async () => {
+      // This matches the exact structure from issue #17318
+      const error = new Error(
+        'request to https://cloudcode-pa.googleapis.com/v1internal:streamGenerateContent failed',
+      );
+      (error as any).type = 'system';
+      (error as any).errno = 'ERR_SSL_SSLV3_ALERT_BAD_RECORD_MAC';
+      (error as any).code = 'ERR_SSL_SSLV3_ALERT_BAD_RECORD_MAC';
+
+      const mockFn = vi
+        .fn()
+        .mockRejectedValueOnce(error)
+        .mockResolvedValue('success');
+
+      const promise = retryWithBackoff(mockFn, {
+        initialDelayMs: 1,
+        maxDelayMs: 1,
+      });
+      await vi.runAllTimersAsync();
+      await expect(promise).resolves.toBe('success');
+      expect(mockFn).toHaveBeenCalledTimes(2);
+    });
   });
 
   describe('Flash model fallback for OAuth users', () => {

packages/core/src/utils/retry.ts

@@ -54,6 +54,12 @@ const RETRYABLE_NETWORK_CODES = [
   'ENOTFOUND',
   'EAI_AGAIN',
   'ECONNREFUSED',
+  // SSL/TLS transient errors
+  'ERR_SSL_SSLV3_ALERT_BAD_RECORD_MAC',
+  'ERR_SSL_WRONG_VERSION_NUMBER',
+  'ERR_SSL_DECRYPTION_FAILED_OR_BAD_RECORD_MAC',
+  'ERR_SSL_BAD_RECORD_MAC',
+  'EPROTO', // Generic protocol error (often SSL-related)
 ];
 
 function getNetworkErrorCode(error: unknown): string | undefined {
@@ -72,8 +78,22 @@ function getNetworkErrorCode(error: unknown): string | undefined {
     return directCode;
   }
 
-  if (typeof error === 'object' && error !== null && 'cause' in error) {
+  // Traverse the cause chain to find error codes (SSL errors are often nested)
-    return getCode((error as { cause: unknown }).cause);
+  let current: unknown = error;
+  const maxDepth = 5; // Prevent infinite loops in case of circular references
+  for (let depth = 0; depth < maxDepth; depth++) {
+    if (
+      typeof current !== 'object' ||
+      current === null ||
+      !('cause' in current)
+    ) {
+      break;
+    }
+    current = (current as { cause: unknown }).cause;
+    const code = getCode(current);
+    if (code) {
+      return code;
+    }
   }
 
   return undefined;