Live session dies with 1011 keepalive ping timeout immediately after first turn_complete (reproducible across gemini-3.1-flash-live-preview, gemini-2.5-flash-native-audio-latest) #2935

Activity

1. himanshu0749

   added

   type: bugError or flaw in code with unintended results or allowing sub-optimal usage patterns.Error or flaw in code with unintended results or allowing sub-optimal usage patterns.

   priority: p2Moderately-important priority. Fix may not be included in next release.Moderately-important priority. Fix may not be included in next release.

   on Sep 3, 2026
2. Venkaiahbabuneelam

   self-assigned this

   on Sep 3, 2026
3. Venkaiahbabuneelam commented on Sep 3, 2026

   

   Venkaiahbabuneelam

   on Sep 3, 2026

   More actions

   Hi @himanshu0749, Thanks for reaching out us!
   To help us understand the issue, could you please share a code snippet of what you've tried so far?
   Thanks
4. Venkaiahbabuneelam

   added

   status:awaiting user response

   on Sep 3, 2026
5. himanshu0749 commented on Sep 5, 2026

   

   himanshu0749

   on Sep 5, 2026

   Author

   More actions

   Sure — here's a minimal repro:

   import os
   import asyncio
   import queue
   import sounddevice as sd
   
   from dotenv import load_dotenv
   from google import genai
   from google.genai import types
   
   load_dotenv()
   
   api_key = os.getenv("GEMINI_API_KEY")
   client = genai.Client(api_key=api_key)
   
   MODEL = "gemini-3.1-flash-live-preview"  # also reproduces on gemini-2.5-flash-native-audio-latest
                                             # and gemini-2.5-flash-native-audio-preview-12-2025
   
   SAMPLE_RATE = 16000
   CHANNELS = 1
   CHUNK_FRAMES = 1600  # 100 ms
   
   
   async def main():
       async with client.aio.live.connect(
           model=MODEL,
           config={
               "response_modalities": ["AUDIO"],
               "input_audio_transcription": {},
           },
       ) as session:
   
           print("Connected")
   
           audio_queue = queue.Queue(maxsize=20)
   
           def mic_callback(indata, frames, time_info, status):
               if status:
                   print(f"Mic status: {status}")
               audio_bytes = indata.copy().tobytes()
               try:
                   audio_queue.put_nowait(audio_bytes)
               except queue.Full:
                   try:
                       audio_queue.get_nowait()
                       audio_queue.put_nowait(audio_bytes)
                   except queue.Empty:
                       pass
   
           async def send_microphone():
               with sd.InputStream(
                   samplerate=SAMPLE_RATE,
                   channels=CHANNELS,
                   dtype="int16",
                   blocksize=CHUNK_FRAMES,
                   callback=mic_callback,
               ):
                   while True:
                       try:
                           audio_bytes = await asyncio.to_thread(audio_queue.get, timeout=0.5)
                       except queue.Empty:
                           continue
   
                       # No gaps observed here before the crash — instrumented
                       # separately with timestamps, sends were continuous.
                       await session.send_realtime_input(
                           audio=types.Blob(
                               data=audio_bytes,
                               mime_type="audio/pcm;rate=16000",
                           )
                       )
   
           async def receive_gemini():
               async for response in session.receive():
                   if not response.server_content:
                       continue
   
                   server_content = response.server_content
   
                   if server_content.input_transcription and server_content.input_transcription.text:
                       print(f"You: {server_content.input_transcription.text}")
   
                   if server_content.model_turn:
                       for part in server_content.model_turn.parts:
                           if part.inline_data:
                               print(f"Gemini audio chunk: {len(part.inline_data.data)} bytes")
   
                   if server_content.turn_complete:
                       print("turn_complete received")
   
           await asyncio.gather(
               send_microphone(),
               receive_gemini(),
           )
   
   
   if __name__ == "__main__":
       asyncio.run(main())

   Running this: speak one short utterance, Gemini replies fully (transcription + all audio chunks arrive, turn_complete prints), then the very next send_realtime_input call in send_microphone() raises:

   websockets.exceptions.ConnectionClosedError: sent 1011 (internal error) keepalive ping timeout; no close frame received
   

   google-genai version: <Name: google-genai Version: 2.22.0 Summary: GenAI Python SDK Home-page: https://github.com/googleapis/python-genai Author: Author-email: Google LLC <[email protected]> License-Expression: Apache-2.0>

   Happy to share a full terminal log if useful.
6. Venkaiahbabuneelam

   removed

   status:awaiting user response

   on Sep 7, 2026
7. kkorpal commented on Sep 8, 2026

   

   kkorpal

   on Sep 8, 2026

   More actions

   Hi @himanshu0749, thanks for the reproduction script! We ran it locally and confirmed the exact behavior: the backend unexpectedly drops the WebSocket with a 1011 error immediately after the first turn_complete. I've escalated this directly to the Live API engineering team to investigate the server-side disconnects. Thanks for flagging this.