First, I will set the hstrs of SRS2 to on by default. As a streaming media server, it should actively fetch content from the origin. Reference: a8781ae

TRANS_BY_GPT3

Analyze the code and find that it is a low-level coroutine object. After actively exiting, the pull function cannot obtain the error.

The upper-level code will start a thread to read messages from the client.

srs_error_t SrsHttpRecvThread::cycle()
{
    srs_error_t err = srs_success;
    
    while ((err = trd->pull()) == srs_success) {
        ISrsHttpMessage* req = NULL;
        SrsAutoFree(ISrsHttpMessage, req);
        
        if ((err = conn->pop_message(&req)) != srs_success) {
            return srs_error_wrap(err, "pop message");

The pop_message function will return an error after the client closes the connection, and then it will be retrieved and returned by the pfn function.

void* SrsSTCoroutine::pfn(void* arg)
{
    SrsSTCoroutine* p = (SrsSTCoroutine*)arg;
    void* res = (void*)p->cycle();
    return res;

And this error code is stored in the return value of the st coroutine.

326	  /* Run thread main */
327	  thread->retval = (*thread->start)(thread->arg);
(gdb) p thread->retval
$1 = (void *) 0x9c87c0

Usually, when the stop function stops the thread, the return value of the coroutine is obtained so that pull can know if there is an error.

void SrsSTCoroutine::stop()
{
    if (!started || disposed) {
        return;
    }
    disposed = true;
    
    interrupt();
    
    void* res = NULL;
    int r0 = st_thread_join((st_thread_t)trd, &res);
    srs_assert(!r0);
    
    // Always override the error by the error from worker.
    if ((srs_error_t)res != srs_success) {
        srs_freep(trd_err);
        trd_err = (srs_error_t)res;

In fact, there is no place to call stop, so pull cannot obtain the error. The solution is to set trd_err when the coroutine naturally exits, so that pull can return the error.

TRANS_BY_GPT3

SRS2, SRS3, Origin, and Edge modes all passed the test without any FD leakage issues.

Fixed.

TRANS_BY_GPT3