Edge: Crash when republish to origin server. When republishing the stream from the edge to the upper layer, a core dump issue occurs occasionally when the socket is closed at the end of the stream.

[mapengfei53]

Description'

Please ensure that the markdown structure is maintained.

When pushing the edge stream to the upper layer, a core dump problem occasionally occurs at the end of the stream.

Please ensure that the markdown structure is maintained.

1. SRS version: srs-3.0release
2. The log of SRS is as follows:

Please ensure that the markdown structure is maintained.

#0  0x00007f672b492337 in raise () from /lib64/libc.so.6
#1  0x00007f672b493a28 in abort () from /lib64/libc.so.6
#2  0x00007f672b48b156 in __assert_fail_base () from /lib64/libc.so.6
#3  0x00007f672b48b202 in __assert_fail () from /lib64/libc.so.6
#4  0x00000000004ad0a7 in srs_close_stfd (stfd=@0x4fc2008: 0x2deaf00) at src/service/srs_service_st.cpp:87
#5  0x00000000004afb9b in SrsTcpClient::close (this=0x4fc2000) at src/service/srs_service_st.cpp:611
#6  0x00000000004af705 in SrsTcpClient::~SrsTcpClient (this=0x4fc2000, __in_chrg=<optimized out>, __vtt_parm=<optimized out>) at src/service/srs_service_st.cpp:581
#7  0x00000000004af992 in SrsTcpClient::~SrsTcpClient (this=0x4fc2000, __in_chrg=<optimized out>, __vtt_parm=<optimized out>) at src/service/srs_service_st.cpp:584
#8  0x00000000004bc1d1 in SrsBasicRtmpClient::close (this=0x4a03d50) at src/service/srs_service_rtmp_conn.cpp:103
#9  0x00000000004bbc99 in SrsBasicRtmpClient::~SrsBasicRtmpClient (this=0x4a03d50, __in_chrg=<optimized out>) at src/service/srs_service_rtmp_conn.cpp:61
#10 0x00000000004cd169 in SrsSimpleRtmpClient::~SrsSimpleRtmpClient (this=0x4a03d50, __in_chrg=<optimized out>) at src/app/srs_app_rtmp_conn.cpp:80
#11 0x00000000004cd198 in SrsSimpleRtmpClient::~SrsSimpleRtmpClient (this=0x4a03d50, __in_chrg=<optimized out>) at src/app/srs_app_rtmp_conn.cpp:82
#12 0x00000000005ba2af in SrsEdgeForwarder::stop (this=0x4225550) at src/app/srs_app_edge.cpp:848
#13 0x00000000005bb43a in SrsPublishEdge::on_proxy_unpublish (this=0x3fb80d0) at src/app/srs_app_edge.cpp:1129
#14 0x00000000004f8b53 in SrsSource::on_edge_proxy_unpublish (this=0x45b99a0) at src/app/srs_app_source.cpp:2940
#15 0x00000000004e416f in SrsRtmpConn::release_publish (this=0x47ff330, source=0x45b99a0) at src/app/srs_app_rtmp_conn.cpp:1498
#16 0x00000000004e0d08 in SrsRtmpConn::publishing (this=0x47ff330, source=0x45b99a0) at src/app/srs_app_rtmp_conn.cpp:1255
#17 0x00000000004dd472 in SrsRtmpConn::stream_service_cycle (this=0x47ff330) at src/app/srs_app_rtmp_conn.cpp:873
#18 0x00000000004dbf63 in SrsRtmpConn::service_cycle (this=0x47ff330) at src/app/srs_app_rtmp_conn.cpp:698
#19 0x00000000004cee70 in SrsRtmpConn::do_cycle (this=0x47ff330) at src/app/srs_app_rtmp_conn.cpp:224
#20 0x00000000004ccd8a in SrsConnection::cycle (this=0x47ff3f8) at src/app/srs_app_conn.cpp:172
#21 0x000000000053194e in SrsSTCoroutine::cycle (this=0x4fb6450) at src/app/srs_app_st.cpp:198
#22 0x00000000005319c3 in SrsSTCoroutine::pfn (arg=0x4fb6450) at src/app/srs_app_st.cpp:213
#23 0x0000000000601bdb in _st_thread_main () at sched.c:337
#24 0x000000000060234b in st_thread_create (start=0xffffffffffffffff, arg=0x4fb6450, joinable=0, stk_size=6296537) at sched.c:616
Backtrace stopped: previous frame inner to this frame (corrupt stack?)

1. The configuration of SRS is as follows:

Please ensure that the markdown structure is maintained.

# the config for srs origin-edge cluster
# @see https://github.com/ossrs/srs/wiki/v1_CN_Edge
# @see full.conf for detail config.

listen              1935;
max_connections     1000;
pid                 objs/edge.pid;
daemon              off;
srs_log_tank        console;
vhost __defaultVhost__ {
    cluster {
        mode            remote;
        origin          127.0.0.1:19350;
    }
}

Current analysis of the reasons:

Please ensure that the markdown structure is maintained.

Currently, the core dump occurs because the function st_netfd_close returns err = -1, which triggers an assert call. Some experts say that this is caused by closing the coroutine before closing the socket.

Please ensure that the markdown structure is maintained.

1. void srs_close_stfd(srs_netfd_t& stfd)
{
    if (stfd) {
        // we must ensure the close is ok.
        int err = st_netfd_close((st_netfd_t)stfd);
        srs_assert(err != -1);
        stfd = NULL;
    }
}
2. void SrsEdgeForwarder::stop()
{
    trd->stop();
    queue->clear();
    srs_freep(sdk);
}
3. The function `SrsEdgeForwarder::stop()` first executes `trd->stop();` and then executes `srs_freep(sdk); srs_freep(sdk);`, which calls the destructor and performs the `close()` operation on the socket. Currently, this issue occurs sporadically, and it is unclear whether this problem is causing it.

Please ensure that the markdown structure is maintained.

Expected behavior (Expect)

Please ensure that the markdown structure is maintained.

Expected to fix this core dump issue.

Please ensure that the markdown structure is maintained.

TRANS_BY_GPT3

[winlinvip]

👍 This is problematic.

TRANS_BY_GPT3

[chen-guanghua]

@mapengfei53, please provide the following information: "Provide the coredump that occurred.

TRANS_BY_GPT3

[winlinvip]

Dup to #1784 has already been updated with the log. We need to wait for the problem to be reproduced, which has a relatively low probability.

Postpone to SRS 5.0.

TRANS_BY_GPT3

[winlinvip]

Dup to #1941