I have a program based on boost::asio and beast with coroutine supports. The program is connecting to a remote server through beast websocket through which it sends requests, which is triggered by its internal logic to the server. So basically the requests are quiet heavy. I use co_spawn(strand_, sendRequest(data), asio::detached) while sendRequest is a coroutine which sends data to the remote server like this ws_.async_write(strand_, asio::buffer(data), asio::use_awaitable), the strand_ at two places are the same one, and the ws_ is initialized with the strand as well:
ssl_context_(asio::ssl::context::tlsv12_client) , ws_(new websocket::stream<beast::ssl_stream<beast::tcp_stream>>(strand_, ssl_context_)), but I got crashed. It happened at soft_mutex.hpp in the beast:

try_lock(T const*)
{
    // If this assert goes off it means you are attempting to
    // simultaneously initiate more than one of same asynchronous
    // operation, which is not allowed. For example, you must wait
    // for an async_read to complete before performing another
    // async_read.
    //
    BOOST_ASSERT(id_ != T::id);
    if(id_ != 0)
        return false;
    id_ = T::id;
    return true;
}

if I replaced co_spawn(strand_, sendRequest(data), asio::detached) with an empty coroutine like this: co_spawn(strand_, []() -> asio::awaitable<void> {co_return;}, asio::detached), it’s fine. So it definitely happens at ws_async_write in the sendRequest. I tried with asio::post to wrap co_spawn, it is still not working. Very annoying. Any help will be highly appreciated!!!

I do this on Ubuntu 22.04 with Boost 1.85