Got no_reply error for my customized protocol about workflow

[Barenboim]

Because client’s main function terminates!
Please: wait_group.wait();
BTW, are you sure your encode() doesn’t have memleak?

[Barenboim]

About exit of the program: https://github.com/sogou/workflow/tree/master/docs/about-exit.md

[Barenboim]

Your implementation of append() cannot garentee you receive the whole message. You know that, right?

[wangh09]

Your implementation of append() cannot garentee you receive the whole message. You know that, right?

Thanks for your reply!
For the append(): yes I know that, just want to see it works for now;
For the enocde(): I have to make the buffer a member of the message right?

[Barenboim]

Your implementation of append() cannot garentee you receive the whole message. You know that, right?

Thanks for your reply!
For the append(): yes I know that, just want to see it works for now;
For the enocde(): I have to make the buffer a member of the message right?

Yes. Make it a member of the message class, allocate memory in it's constructor or encode() function and release in it's destructor. But don't do this:

        using NTF = WFNetworkTaskFactory<protocol::OrderMessage, protocol::OrderMessage>;
        WFTestTask *task = NTF::create_client_task(TT_TCP, host, port,
                                                       retry_max,
                                                       std::move(callback));
        task->set_keep_alive(30 * 1000);
        task->get_resp()->buffer = new char[MAX];     // not very good.
        return task;

This makes the response class non-RAII. But the response object may be destroyed and reconstructed during communication when retrial occurs. If it's not a RAII class, you have to set prepare function or make a derivation of WFNetworkTask.

[wangh09]

Another question about the client: I see it is tasks that can be kept alive, but a task seems to be not thread safe. Then how to reuse the connections?
Say my architecture is Client <-> ServerA <-> ServerB, ServerA accepts stateless http requests from the Client, and passes them to ServerB on persistent tcp connections (10 for example). should I create 10 tasks and use locks to share them?

[Barenboim]

So you are writing a ServerA? Please refer to tutorial-05-http_proxy
Note: connections can be reused, but tasks cannot be reused by users.

[Barenboim]

task->set_keep_alive() sets the connection underneath to be reused or not, not the task.
For http protocol, adding http header “Connection” is a better way. Though both ways work.