Closed matt-mcallister closed 2 weeks ago
I suppose that I am suffering from the same problem. https://stackoverflow.com/questions/78337450/publishing-message-to-gcp-pubsub-from-a-fastapi-app is there any workaround?
As I am using the PublisherClient from CloudRun, I can´t find way to reuse the same PublisherClient. It seems a so basic use case.
Hello,
I have exactly the same problem. I use the PublisherClient
in an Apache Beam (Dataflow) job and I noticed mutual blocking problems.
I have a fairly simple way to reproduce it:
import threading
from google.cloud import pubsub_v1
def pubsub():
for i in [1, 2, 3, 4, 5]:
print(i)
pubsub_v1.PublisherClient()
t1 = threading.Thread(target=pubsub)
t2 = threading.Thread(target=pubsub)
t1.start()
t2.start()
t1.join()
t2.join()
Producing:
$ python block.py
1
1
^C^C^C^C
[1] 63732 terminated python3 block.py
OS: macOS 14.4.1 (23E224) - M2 Pro
Python: Python 3.12.1
google-cloud-pubsub: 2.21.1
I hope this will help you. If there is a need to open another issue, no problem.
Thank you in advance for your help.
Hi, is there a reason you would want to create multiple client library instances per thread, instead of creating a single client library instance and re-using it across multiple threads?
As I am using the PublisherClient from CloudRun, I can´t find way to reuse the same PublisherClient. It seems a so basic use case.
When the Cloud Run instance instantiates, could you create a single instance of the client library, and re-use it across the lifetime of the Cloud run container instance?
I don't have the setup to test PublisherClient with credentials so I changed run
to the following:
def run(publisher_number):
# PublisherClient.from_service_account_info(creds)
PublisherClient()
print(f'Created publisher {publisher_number}')
Using code example, I'm not able to reproduce the issue in python:3.10.13-bookworm
docker image, maybe it's fixed in latest version?
Tried again just for grpc.secure_channel
, I changed our example so that I can test creating multiple secure channel in different thread using composite_channel_credentials
similar to what PublisherClient and python-api-core is doing:
import argparse
import contextlib
import logging
import concurrent
import _credentials
import grpc
helloworld_pb2, helloworld_pb2_grpc = grpc.protos_and_services(
"helloworld.proto"
)
_LOGGER = logging.getLogger(__name__)
_LOGGER.setLevel(logging.INFO)
_SERVER_ADDR_TEMPLATE = "localhost:%d"
@contextlib.contextmanager
def create_client_channel(addr):
# Call credential object will be invoked for every single RPC
call_credentials = grpc.access_token_call_credentials(
"example_oauth2_token"
)
# Channel credential will be valid for the entire channel
channel_credential = grpc.ssl_channel_credentials(
_credentials.ROOT_CERTIFICATE
)
# Combining channel credentials and call credentials together
composite_credentials = grpc.composite_channel_credentials(
channel_credential,
call_credentials,
)
channel = grpc.secure_channel(addr, composite_credentials)
yield channel
def send_rpc(channel):
stub = helloworld_pb2_grpc.GreeterStub(channel)
request = helloworld_pb2.HelloRequest(name="you")
try:
response = stub.SayHello(request)
except grpc.RpcError as rpc_error:
_LOGGER.error("Received error: %s", rpc_error)
return rpc_error
else:
_LOGGER.info("Received message: %s", response)
return response
def run(port):
with create_client_channel(_SERVER_ADDR_TEMPLATE % port) as channel:
send_rpc(channel)
def main():
parser = argparse.ArgumentParser()
parser.add_argument(
"--port",
nargs="?",
type=int,
default=50051,
help="the address of server",
)
args = parser.parse_args()
with concurrent.futures.ThreadPoolExecutor(max_workers=2) as executor:
total_publishers = 10
futures = []
for i in range(0, total_publishers):
futures.append(executor.submit(run, args.port))
print(f"Submitted all {total_publishers} to thread executor")
for f in futures:
f.result()
if __name__ == "__main__":
logging.basicConfig(level=logging.INFO)
main()
And still not able to reproduce.
Closing due to no response. Feel free to reopen.
Environment details
python:3.10.13-bookworm
as the base image3.10.13
23.0.1
google-cloud-pubsub
version:2.19.7
Steps to reproduce
Calling
PublisherClient.from_service_account_info()
from multiple concurrent threads causes the program to hang.Code example
Output
Stack trace
The program hangs so there is no stack trace, but this is the core dump