During testing the behavior of some software components when network is turned off I experienced a deadlock during the ssl_stream::async_handshake of boost.beast. I did only experience this deadlock with running Zscaler client connector and did not test if this issue also occurs using other VPN's. Below a minimal compiling program, the log output, and a step-by-step guide on how to reproduce the deadlock. Let me know if you experience similar issues or if I am doing something wrong. Thanks in advance for your input and support!
Version of Beast and Asio
boost.asio: 1.28.0
boost.beast: 347
Steps necessary to reproduce the problem
See here for a minimal compiling program and see deadlock.log for the full program output with BOOST_ASIO_ENABLE_HANDLER_TRACKING enabled.
The following steps reproduce the deadlock on my machine:
Start Zscaler Client Connector.
Start program with BOOST_ASIO_ENABLE_HANDLER_TRACKING=ON
Turn Off network on OS level.
Wait until program gets stuck and does not connect anymore.
Use boost::asio::use_future for the async_connect and async_handshake calls and detach the io_context thread if the future does not return after e.g. 30s. However, this is no proper solution and does leak memory. It would be much better to tackle this deadlock on library level and prevent it from happening at all.
During testing the behavior of some software components when network is turned off I experienced a deadlock during the ssl_stream::async_handshake of boost.beast. I did only experience this deadlock with running Zscaler client connector and did not test if this issue also occurs using other VPN's. Below a minimal compiling program, the log output, and a step-by-step guide on how to reproduce the deadlock. Let me know if you experience similar issues or if I am doing something wrong. Thanks in advance for your input and support!
Version of Beast and Asio
boost.asio: 1.28.0 boost.beast: 347
Steps necessary to reproduce the problem
See here for a minimal compiling program and see deadlock.log for the full program output with BOOST_ASIO_ENABLE_HANDLER_TRACKING enabled.
The following steps reproduce the deadlock on my machine:
@asio|1688107293.822678|0|resolver@0x7fffb42679d8.canceland the program does not recover (at least not within 20min). Stepping in with a debugger shows that the io_context thread is stuck at the following line: https://github.com/chriskohlhoff/asio/blob/c465349fa5cd91a64bb369f5131ceacab2c0c1c3/asio/include/asio/detail/impl/epoll_reactor.ipp#L509All relevant compiler information
OS: Ubuntu 22.04.1 GCC: 11.3.0 CMake: 3.25.2 Ninja: 1.10.2
Possible Workaround
Use
boost::asio::use_futurefor the async_connect and async_handshake calls and detach the io_context thread if the future does not return after e.g. 30s. However, this is no proper solution and does leak memory. It would be much better to tackle this deadlock on library level and prevent it from happening at all.