Skip to navigation

Streaming

View as Markdown

gRPC supports four types of service methods: unary, server streaming, client streaming, and bidirectional streaming. Streaming enables efficient real-time communication and large data transfers.

Server streaming

Server streaming allows the server to send multiple responses to a single client request:

streaming_service.proto
syntax = "proto3";
package streaming.v1;
service StreamingService {
// Server streaming: send multiple events to client
rpc StreamUserEvents(StreamUserEventsRequest) returns (stream UserEvent);
// Server streaming: download large file in chunks
rpc DownloadFile(DownloadFileRequest) returns (stream FileChunk);
// Server streaming: real-time notifications
rpc SubscribeToNotifications(SubscribeRequest) returns (stream Notification);
}
message StreamUserEventsRequest {
string user_id = 1;
repeated UserEventType event_types = 2;
google.protobuf.Timestamp start_time = 3;
}
message UserEvent {
string id = 1;
string user_id = 2;
UserEventType type = 3;
google.protobuf.Timestamp timestamp = 4;
google.protobuf.Any data = 5;
}
enum UserEventType {
USER_EVENT_TYPE_UNSPECIFIED = 0;
USER_EVENT_TYPE_LOGIN = 1;
USER_EVENT_TYPE_LOGOUT = 2;
USER_EVENT_TYPE_PROFILE_UPDATE = 3;
USER_EVENT_TYPE_PASSWORD_CHANGE = 4;
}

Server streaming implementation:

server_streaming.py
import grpc
import time
import asyncio
from grpc import ServicerContext
from typing import Iterator
class StreamingServiceServicer(streaming_pb2_grpc.StreamingServiceServicer):
def StreamUserEvents(
self,
request: streaming_pb2.StreamUserEventsRequest,
context: ServicerContext
) -> Iterator[streaming_pb2.UserEvent]:
"""Stream user events in real-time."""
# Validate request
if not request.user_id:
context.set_code(grpc.StatusCode.INVALID_ARGUMENT)
context.set_details('User ID is required')
return
# Subscribe to event stream
event_subscriber = self.event_store.subscribe(
user_id=request.user_id,
event_types=request.event_types,
start_time=request.start_time
)
try:
while context.is_active():
# Wait for next event (with timeout)
try:
event = event_subscriber.get_next_event(timeout=30)
if event:
yield event
else:
# Send heartbeat to keep connection alive
continue
except TimeoutError:
# Check if client is still connected
if not context.is_active():
break
except Exception as e:
context.set_code(grpc.StatusCode.INTERNAL)
context.set_details(f'Error streaming events: {str(e)}')
break
finally:
# Clean up subscription
event_subscriber.close()
def DownloadFile(
self,
request: streaming_pb2.DownloadFileRequest,
context: ServicerContext
) -> Iterator[streaming_pb2.FileChunk]:
"""Download file in chunks."""
try:
file_path = self.file_store.get_file_path(request.file_id)
if not file_path or not os.path.exists(file_path):
context.set_code(grpc.StatusCode.NOT_FOUND)
context.set_details(f'File {request.file_id} not found')
return
chunk_size = 64 * 1024 # 64KB chunks
with open(file_path, 'rb') as f:
while True:
chunk_data = f.read(chunk_size)
if not chunk_data:
break
# Check if client disconnected
if not context.is_active():
break
yield streaming_pb2.FileChunk(
data=chunk_data,
offset=f.tell() - len(chunk_data),
size=len(chunk_data)
)
except Exception as e:
context.set_code(grpc.StatusCode.INTERNAL)
context.set_details(f'Error downloading file: {str(e)}')

Client streaming

Client streaming allows the client to send multiple requests and receive a single response:

client_streaming.proto
service StreamingService {
// Client streaming: upload large file in chunks
rpc UploadFile(stream FileChunk) returns (UploadFileResponse);
// Client streaming: batch data processing
rpc ProcessDataBatch(stream DataRecord) returns (BatchProcessingResult);
// Client streaming: real-time metrics collection
rpc CollectMetrics(stream MetricData) returns (MetricsCollectionResult);
}
message FileChunk {
oneof data {
FileMetadata metadata = 1;
bytes chunk = 2;
}
}
message FileMetadata {
string filename = 1;
int64 total_size = 2;
string content_type = 3;
string checksum = 4;
}
message UploadFileResponse {
string file_id = 1;
int64 bytes_uploaded = 2;
string download_url = 3;
bool checksum_verified = 4;
}

Client streaming implementation:

client_streaming.py
class StreamingServiceServicer(streaming_pb2_grpc.StreamingServiceServicer):
def UploadFile(
self,
request_iterator: Iterator[streaming_pb2.FileChunk],
context: ServicerContext
) -> streaming_pb2.UploadFileResponse:
"""Upload file from client stream."""
file_metadata = None
total_bytes = 0
file_path = None
hasher = hashlib.sha256()
try:
for chunk in request_iterator:
if chunk.HasField('metadata'):
# First chunk contains metadata
file_metadata = chunk.metadata
# Create temporary file
file_id = str(uuid.uuid4())
file_path = f'/tmp/uploads/{file_id}'
os.makedirs(os.path.dirname(file_path), exist_ok=True)
elif chunk.HasField('chunk'):
# Subsequent chunks contain file data
if not file_metadata:
context.set_code(grpc.StatusCode.INVALID_ARGUMENT)
context.set_details('File metadata must be sent first')
return streaming_pb2.UploadFileResponse()
# Write chunk to file
with open(file_path, 'ab') as f:
f.write(chunk.chunk)
total_bytes += len(chunk.chunk)
hasher.update(chunk.chunk)
# Check size limit
if total_bytes > file_metadata.total_size:
context.set_code(grpc.StatusCode.INVALID_ARGUMENT)
context.set_details('File size exceeds declared size')
return streaming_pb2.UploadFileResponse()
# Verify checksum
computed_checksum = hasher.hexdigest()
checksum_verified = computed_checksum == file_metadata.checksum
if not checksum_verified:
context.set_code(grpc.StatusCode.DATA_LOSS)
context.set_details('File checksum verification failed')
return streaming_pb2.UploadFileResponse()
# Move file to permanent storage
permanent_path = self.file_store.store_file(file_id, file_path)
download_url = self.file_store.get_download_url(file_id)
return streaming_pb2.UploadFileResponse(
file_id=file_id,
bytes_uploaded=total_bytes,
download_url=download_url,
checksum_verified=True
)
except Exception as e:
context.set_code(grpc.StatusCode.INTERNAL)
context.set_details(f'Error uploading file: {str(e)}')
return streaming_pb2.UploadFileResponse()
finally:
# Clean up temporary file
if file_path and os.path.exists(file_path):
os.remove(file_path)

Bidirectional streaming

Bidirectional streaming allows both client and server to send multiple messages:

bidirectional_streaming.proto
service StreamingService {
// Bidirectional streaming: real-time chat
rpc Chat(stream ChatMessage) returns (stream ChatMessage);
// Bidirectional streaming: real-time collaboration
rpc Collaborate(stream CollaborationEvent) returns (stream CollaborationEvent);
// Bidirectional streaming: live data processing
rpc ProcessLiveData(stream DataInput) returns (stream ProcessingResult);
}
message ChatMessage {
string id = 1;
string user_id = 2;
string room_id = 3;
string content = 4;
google.protobuf.Timestamp timestamp = 5;
ChatMessageType type = 6;
}
enum ChatMessageType {
CHAT_MESSAGE_TYPE_UNSPECIFIED = 0;
CHAT_MESSAGE_TYPE_TEXT = 1;
CHAT_MESSAGE_TYPE_IMAGE = 2;
CHAT_MESSAGE_TYPE_FILE = 3;
CHAT_MESSAGE_TYPE_SYSTEM = 4;
CHAT_MESSAGE_TYPE_TYPING = 5;
}

Bidirectional streaming implementation:

bidirectional_streaming.py
import asyncio
import queue
import threading
class StreamingServiceServicer(streaming_pb2_grpc.StreamingServiceServicer):
def Chat(
self,
request_iterator: Iterator[streaming_pb2.ChatMessage],
context: ServicerContext
) -> Iterator[streaming_pb2.ChatMessage]:
"""Bidirectional chat streaming."""
# Message queue for outgoing messages
outgoing_queue = queue.Queue()
# Track user session
user_session = None
def handle_incoming_messages():
"""Process incoming messages from client."""
nonlocal user_session
try:
for message in request_iterator:
if not user_session:
# First message establishes session
user_session = self.chat_service.join_room(
user_id=message.user_id,
room_id=message.room_id
)
# Send welcome message
welcome_msg = streaming_pb2.ChatMessage(
id=str(uuid.uuid4()),
user_id="system",
room_id=message.room_id,
content=f"User {message.user_id} joined the chat",
timestamp=google.protobuf.timestamp_pb2.Timestamp(),
type=streaming_pb2.CHAT_MESSAGE_TYPE_SYSTEM
)
outgoing_queue.put(welcome_msg)
# Process message
if message.type == streaming_pb2.CHAT_MESSAGE_TYPE_TEXT:
# Broadcast to other users in room
self.chat_service.broadcast_message(message)
elif message.type == streaming_pb2.CHAT_MESSAGE_TYPE_TYPING:
# Send typing indicator to other users
self.chat_service.broadcast_typing(message)
except Exception as e:
print(f"Error handling incoming messages: {e}")
finally:
# Clean up session
if user_session:
self.chat_service.leave_room(user_session)
# Start background thread for incoming messages
incoming_thread = threading.Thread(target=handle_incoming_messages)
incoming_thread.daemon = True
incoming_thread.start()
# Subscribe to room messages
message_subscriber = None
if user_session:
message_subscriber = self.chat_service.subscribe_to_room(
user_session.room_id,
exclude_user=user_session.user_id
)
try:
while context.is_active():
# Check for outgoing messages from queue
try:
message = outgoing_queue.get(timeout=1)
yield message
except queue.Empty:
pass
# Check for messages from other users
if message_subscriber:
try:
room_message = message_subscriber.get_message(timeout=1)
if room_message:
yield room_message
except TimeoutError:
pass
finally:
# Clean up
if message_subscriber:
message_subscriber.close()
if user_session:
self.chat_service.leave_room(user_session)

Streaming best practices

Flow control

Implement proper flow control to prevent overwhelming clients:

flow_control.py
def StreamData(self, request, context):
"""Stream with flow control."""
# Use a bounded queue to control memory usage
data_queue = queue.Queue(maxsize=100)
def data_producer():
"""Background thread that produces data."""
for item in self.data_source.get_items():
try:
data_queue.put(item, timeout=5)
except queue.Full:
# Apply backpressure
print("Client is too slow, dropping data")
break
producer_thread = threading.Thread(target=data_producer)
producer_thread.start()
try:
while context.is_active():
try:
item = data_queue.get(timeout=30)
yield item
except queue.Empty:
# Send heartbeat or check client connection
if not context.is_active():
break
finally:
producer_thread.join(timeout=1)

Error handling in streams

Handle errors gracefully in streaming operations:

stream_error_handling.py
def StreamWithErrorHandling(self, request, context):
"""Stream with robust error handling."""
try:
for item in self.get_stream_data(request):
if not context.is_active():
break
try:
# Process item
processed_item = self.process_item(item)
yield processed_item
except ProcessingError as e:
# Send error as part of response
error_response = create_error_response(e)
yield error_response
except Exception as e:
# Critical error - abort stream
context.set_code(grpc.StatusCode.INTERNAL)
context.set_details(f'Processing failed: {str(e)}')
break
except Exception as e:
context.set_code(grpc.StatusCode.INTERNAL)
context.set_details(f'Stream failed: {str(e)}')

Client-side streaming

Handle streaming on the client side:

streaming_client.py
import grpc
def stream_chat_client():
"""Example bidirectional streaming client."""
channel = grpc.insecure_channel('localhost:50051')
stub = streaming_pb2_grpc.StreamingServiceStub(channel)
def message_generator():
"""Generate outgoing messages."""
# Send initial message
yield streaming_pb2.ChatMessage(
user_id="user123",
room_id="general",
content="Hello, world!",
type=streaming_pb2.CHAT_MESSAGE_TYPE_TEXT
)
# Keep connection alive and send periodic messages
while True:
user_input = input("Enter message: ")
if user_input.lower() == 'quit':
break
yield streaming_pb2.ChatMessage(
user_id="user123",
room_id="general",
content=user_input,
type=streaming_pb2.CHAT_MESSAGE_TYPE_TEXT
)
# Start bidirectional stream
responses = stub.Chat(message_generator())
try:
for response in responses:
print(f"Received: {response.content}")
except grpc.RpcError as e:
print(f"RPC failed: {e}")

Streaming in gRPC enables powerful real-time applications while maintaining the benefits of strongly-typed contracts and efficient binary protocols.