What is the proper way to shutdown a session?

[mangodan2003]

I am using only pub so far.
The basic examples all (zenoh-c and zenonh-cpp) seem to shutdown cleanly but as soon as I try to integrate into a real application I am hitting rust panics or just hangs at exit.

I am using zenoh-c and zenoh-cpp 1.0.0.5

This is the simplest example that reproduces the panic - note, holding the session within a class which is destructed causes the panic. I have tried using pure zenoh-c and calling z_close in the class destructor but thus doesn't help. In fact not calling z_close stops the panic and in a simple example makes everything appear good, but in our real world app something hangs and prevents the app from exiting.

#include <memory>
#include "zenoh.hxx"
using namespace zenoh;

const char* keyexpr = "test";

class ZenohPub {
public:
  static ZenohPub& getZenohPub() {
    static ZenohPub instance;
    return instance;
  }

  void send(const char* buf) {
    pub->put(Bytes::serialize(buf), {.encoding = Encoding("text/plain")});
  }

private:
  ZenohPub() {
    auto config = Config::create_default();
    session = std::make_unique<zenoh::Session>(Session::open(std::move(config)));
    pub = std::make_unique<zenoh::Publisher>(session->declare_publisher(KeyExpr(keyexpr)));
  }
  std::unique_ptr<zenoh::Session> session;
  std::unique_ptr<zenoh::Publisher> pub;
};

int main(int argc, char **argv) {
  ZenohPub& zp = ZenohPub::getZenohPub();
  zp.send("test!");
}

Build with:
g++ -o test test.cpp -I zenoh-cpp/include/ -I zenoh-c/include/ -DZENOHCXX_ZENOHC -lzenohc -L zenoh-c/target/release

Run with:
LD_LIBRARY_PATH=zenoh-c/target/release ./test

The output of which is:
thread '<unnamed>' panicked at /rustc/129f3b9964af4d4a709d1383930ade12dfe7c081/library/std/src/thread/local.rs:262:26: cannot access a Thread Local Storage value during or after destruction: AccessError note: run with RUST_BACKTRACE=1environment variable to display a backtrace Aborted (core dumped)

Many thanks!

[wyfo]

The difference between the zenoh examples and your code is the keyword static. In fact, according to the specifications, static objects are destructed in atexit.
However, the Rust layer behind the C/C++ API doesn't work well in atexit context; Rust thread local storage has indeed already been destructed at that time, and z_close panics when accessing it.
This is a known issue, and the only solution we can advise for now is to avoid calling z_close (or Session destructor) in atexit, thus avoid static session object.