Goose-Bomb opened a new issue #8937:
URL: https://github.com/apache/tvm/issues/8937
As long as I use the IPv6 address as the host, the C++ RPC program will
produce an **Address family not supported by protocol** error.
To reproduce this error, run:
`./tvm_rpc server --host=:: --port=9090`, this error will occur:
logs:
```log
[20:44:08] /home/nullko/Documents/tvm-vsi_npu/apps/cpp_rpc/main.cc:96: host
= ::
[20:44:08] /home/nullko/Documents/tvm-vsi_npu/apps/cpp_rpc/main.cc:97: port
= 9090
[20:44:08] /home/nullko/Documents/tvm-vsi_npu/apps/cpp_rpc/main.cc:98:
port_end = 9099
[20:44:08] /home/nullko/Documents/tvm-vsi_npu/apps/cpp_rpc/main.cc:99:
tracker =
[20:44:08] /home/nullko/Documents/tvm-vsi_npu/apps/cpp_rpc/main.cc:100: key
=
[20:44:08] /home/nullko/Documents/tvm-vsi_npu/apps/cpp_rpc/main.cc:101:
custom_addr =
[20:44:08] /home/nullko/Documents/tvm-vsi_npu/apps/cpp_rpc/main.cc:102:
work_dir =
[20:44:08] /home/nullko/Documents/tvm-vsi_npu/apps/cpp_rpc/main.cc:103:
silent = False
[20:44:08] /home/nullko/Documents/tvm-vsi_npu/apps/cpp_rpc/main.cc:264:
Starting CPP Server, Press Ctrl+C to stop.
[20:44:08]
/home/nullko/Documents/tvm-vsi_npu/apps/cpp_rpc/../../src/support/socket.h:265:
Warning: Bind failed to :::9090
terminate called after throwing an instance of 'tvm::runtime::InternalError'
what(): [20:44:08]
/home/nullko/Documents/tvm-vsi_npu/apps/cpp_rpc/../../src/support/socket.h:362:
Socket TryBindHost Error:Address family not supported by protocol
Stack trace:
[bt] (0)
/usr/local/lib/libtvm_runtime.so(tvm::runtime::Backtrace[abi:cxx11]()+0x10)
[0xb6ea2f34]
[bt] (1) ./tvm_rpc(tvm::runtime::detail::LogFatal::Entry::Finalize()+0xcc)
[0x4c8204]
[bt] (2) ./tvm_rpc(tvm::support::Socket::Error(char const*)+0xf8)
[0x4d14c8]
[bt] (3)
./tvm_rpc(tvm::support::Socket::TryBindHost(std::__cxx11::basic_string<char,
std::char_traits<char>, std::allocator<char> >, int, int)+0x170) [0x4d0010]
[bt] (4) ./tvm_rpc(tvm::runtime::RPCServer::Start()+0xe0) [0x4cfa24]
[bt] (5)
./tvm_rpc(tvm::runtime::RPCServerCreate(std::__cxx11::basic_string<char,
std::char_traits<char>, std::allocator<char> >, int, int,
std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>
>, std::__cxx11::basic_string<char, std::char_traits<char>,
std::allocator<char> >, std::__cxx11::basic_string<char,
std::char_traits<char>, std::allocator<char> >,
std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char>
>, bool)+0x33c) [0x4cf66c]
[bt] (6) ./tvm_rpc(RpcServer(int, char**)+0x488) [0x4c7364]
[bt] (7) ./tvm_rpc(main+0x1bc) [0x4c772c]
[bt] (8) /lib/arm-linux-gnueabihf/libc.so.6(__libc_start_main+0x99)
[0xb6b3c066]
```
It seems that the cause is when the socket is created, it by default uses
`PF_INET` protocol, without checking what is the actual protocol of the given
IP address, as in
https://github.com/apache/tvm/blob/7eda4a5342851c9938125086e027d8d1bd5ca298/apps/cpp_rpc/rpc_server.cc#L128
A quick but ugly workaround could be like this:
```c++
void Start() {
support::SockAddr addr(host_.c_str(), 0);
listen_sock_.Create(addr.ss_family());
my_port_ = listen_sock_.TryBindHost(host_, port_, port_end_);
LOG(INFO) << "bind to " << host_ << ":" << my_port_;
listen_sock_.Listen(1);
std::future<void> proc(std::async(std::launch::async,
&RPCServer::ListenLoopProc, this));
proc.get();
// Close the listen socket
listen_sock_.Close();
}
```
Or maybe the socket wrapper could be tweaked to support both IPv4 and IPv6
addresses at the same time, see this post:
https://www.ibm.com/docs/en/i/7.2?topic=sscaaiic-example-accepting-connections-from-both-ipv6-ipv4-clients
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
