[ARROW-11866] [C++] Arrow Flight SetShutdownOnSignals cause potential mutex deadlock in gRPC

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 3.0.0
• Fix Version/s: 4.0.0
• Component/s: FlightRPC
• Environment: * Arrow 3.0 ; * gcc (Debian 10.2.0-16) 10.2.0 ; * grpc 1.33.1/C++
• External issue URL: https://github.com/apache/arrow/issues/27711
• Assignee: Unassigned
• Reporter: Lynch Wu
• Created: 05/Mar/21 00:58
• Updated: 11/Jan/23 08:22
• Resolved: 05/Mar/21 01:30

Description

1. When starting a Flight Server with shut down signal (SIGTERM) and using `Ctrl + C` to terminate the server cause potential mutex deadlock in gRPC.

• gcc (Debian 10.2.0-16) 10.2.0
• grpc 1.33.1/C++

2. https://github.com/apache/arrow/blob/master/cpp/src/arrow/flight/test_server.cc This test case can be used as minimal code to re-produce the issue with above gRPC version and gcc version on Debian 10.2.0.

3. Related gRPC issue is: https://github.com/grpc/grpc/issues/24884

4. I am wandering whether the way handling the signal is proper with gRPC here: https://github.com/apache/arrow/blob/master/cpp/src/arrow/flight/server.cc#L881

5. The recommended way to use a completion queue: https://github.com/grpc/grpc/blob/v1.35.0/examples/cpp/helloworld/greeter_async_server.cc

 

Feel free to share your thoughts for this.

Issue Links

• duplicates: ARROW-11560 [FlightRPC][C++][Python] Interrupting a Flight server results in abort