fix hls & dash segments cleanup.

[suzp1984]

Impacted Version

Both v6 and v7.

How to reproduce?

1. go to develop or 6.0release branch.
2. compiling srs
3. ./objs/srs -c conf/hls.conf
4. ffmpeg -re -i ./doc/source.flv -c copy -f flv rtmp://localhost/live/livestream publish a stream for more than 10 seconds to generate the .m3u8 and .ts file.
5. type q to quite ffmpeg push stream.
6. check ./objs/nginx/html/live/ folder, whether the .m3u8 and .ts files still exist.

the expected behavior is that the .m3u8 and .ts files should cleanup a few seconds later after rtmp stop publish.

Cause

Introduced by #4097

The hls segments cleanup process is running in this way previously.

srs/trunk/src/app/srs_app_source.cpp

Lines 1847 to 1875 in f8319d6

	srs_error_t SrsLiveSourceManager::notify(int event, srs_utime_t interval, srs_utime_t tick)
	{
	srs_error_t err = srs_success;
	std::map< std::string, SrsSharedPtr<SrsLiveSource> >::iterator it;
	for (it = pool.begin(); it != pool.end();) {
	SrsSharedPtr<SrsLiveSource>& source = it->second;
	// Do cycle source to cleanup components, such as hls dispose.
	if ((err = source->cycle()) != srs_success) {
	SrsContextId cid = source->source_id();
	if (cid.empty()) cid = source->pre_source_id();
	return srs_error_wrap(err, "source cycle, id=[%s]", cid.c_str());
	}
	// When source expired, remove it.
	// @see https://github.com/ossrs/srs/issues/713
	if (source->stream_is_dead()) {
	SrsContextId cid = source->source_id();
	if (cid.empty()) cid = source->pre_source_id();
	srs_trace("Live: cleanup die source, id=[%s], total=%d", cid.c_str(), (int)pool.size());
	pool.erase(it++);
	} else {
	++it;
	}
	}
	return err;
	}

A timer run above method in period, source->cycle() will run the hls and dash dispose() method to cleanup segment files.
But the cycle() will exit before run dispose, after #4097 introduced the enabled variable.

    if (!enabled) {
        return err;
    }

Solution

run source->dispose(); before destroy SrsLiveSource.

[winlinvip]

In the source cycle, it will call hls dispose:

srs_error_t SrsLiveSourceManager::notify(int event, srs_utime_t interval, srs_utime_t tick) {
        // Do cycle source to cleanup components, such as hls dispose.
        if ((err = source->cycle()) != srs_success) {

srs_error_t SrsOriginHub::cycle() {
    if ((err = hls->cycle()) != srs_success) {
        return srs_error_wrap(err, "hls cycle");
    }

srs_error_t SrsHls::cycle() {
    if (srs_get_system_time() - last_update_time <= hls_dispose) {
        return err;
    }

    dispose();

The stream dead detection will check the delay cleanup:

srs_error_t SrsLiveSourceManager::notify(int event, srs_utime_t interval, srs_utime_t tick)
        if (source->stream_is_dead()) {

bool SrsLiveSource::stream_is_dead() {
    if (now < stream_die_at_ + hub->cleanup_delay()) {
        return false;
    }

srs_utime_t SrsOriginHub::cleanup_delay() {
    srs_utime_t hls_delay = hls->cleanup_delay();

srs_utime_t SrsHls::cleanup_delay()
{
    if (!enabled) {
        return 0;
    }

    // We use larger timeout to cleanup the HLS, after disposed it if required.
    return _srs_config->get_hls_dispose(req->vhost) * 1.1;
}

After FFmepg quit, the HLS is not enabled, so it also disable the dispose. I think we should fix this issue by modify SrsHls::cleanup_delay rather than call hls dispose again.

In general, we should think about how to dispose a !enabled stream source in cycle, not dispose it again before destroy it.

[cdgraff]

+1 I reach here for the same issue, some timeline to fix this? or merge this changes?

[pisken]

hi, how long to release this fix? I'm having this problem and it's causing disk space overflow.