fix #244, conn thread use cond to wait for recv thread error. 2.0.47.

ossrs · Dec 3, 2014 · 787ab67 · 787ab67
1 parent 565f29e
commit 787ab67
Show file tree

Hide file tree

Showing 4 changed files with 51 additions and 39 deletions.
diff --git a/README.md b/README.md
@@ -485,6 +485,7 @@ Supported operating systems and hardware:
 * 2013-10-17, Created.<br/>
 
 ## History
+* v2.0, 2014-12-03, fix [#244](https://github.com/winlinvip/simple-rtmp-server/issues/244), conn thread use cond to wait for recv thread error. 2.0.47.
 * v2.0, 2014-12-02, merge [#239](https://github.com/winlinvip/simple-rtmp-server/pull/239), traverse the token before response connect. 2.0.45.
 * v2.0, 2014-12-02, srs-librtmp support hijack io apis for st-load. 2.0.42.
 * v2.0, 2014-12-01, for [#237](https://github.com/winlinvip/simple-rtmp-server/issues/237), refine syscall for recv, supports 1.5k clients. 2.0.41.
@@ -703,47 +704,30 @@ Supported operating systems and hardware:
 
 ## Performance
 
-Performance benchmark history, on virtual box:
+Performance benchmark history, on virtual box.
+
+### Play benchmark
+
+The play benchmark by st-load:
 
 * 2013-11-28, SRS 0.5.0,   1.8k(1800)clients, 90%CPU, 41MB. [benchmark](https://github.com/winlinvip/simple-rtmp-server/commit/023e23bc8261bec15a70a7ae932098fb4f82b679)
 * 2014-07-12, SRS 0.9.156, 1.8k(1800)clients, 68%CPU, 38MB. [benchmark](https://github.com/winlinvip/simple-rtmp-server/commit/e2d273f4939348374bf9644df9d54c4293b39c1a)
 * 2014-07-12, SRS 0.9.156, 2.7k(2700)clients, 89%CPU, 61MB. [benchmark](https://github.com/winlinvip/simple-rtmp-server/commit/6d12280b7cc54c465b1caf8b1402149e77c4c7d9)
-* 2014-11-11, SRS 1.0.5,   2.7k(2700)clients, 85%CPU, 66MB. (1.0 equals 2.0.12)
+* 2014-11-11, SRS 1.0.5,   2.7k(2700)clients, 85%CPU, 66MB.
+* 2014-11-11, SRS 2.0.12,  2.7k(2700)clients, 85%CPU, 66MB.
 * 2014-11-12, SRS 2.0.14,  2.7k(2700)clients, 69%CPU, 59MB.
 * 2014-11-12, SRS 2.0.14,  3.5k(3500)clients, 95%CPU, 78MB.
-* 2014-11-13, SRS 2.0.15,  6.0k(6000)clients, 82%CPU, 203MB. (500 publishers).
+* 2014-11-13, SRS 2.0.15,  6.0k(6000)clients, 82%CPU, 203MB.
 * 2014-11-22, SRS 2.0.30,  7.5k(7500)clients, 87%CPU, 320MB. 
-* 2014-12-01, SRS 2.0.41,  7.5k(7500)clients, 87%CPU, 320MB. (1500 publishers).
 
-Latest benchmark(2014-07-12):
+* See also: [Performance for x86/x64 Test Guide](https://github.com/winlinvip/simple-rtmp-server/wiki/v1_CN_Performance)
+* See also: [Performance for RaspberryPi](https://github.com/winlinvip/simple-rtmp-server/wiki/v1_CN_RaspberryPi)
 
-1.  300 connections,  150Mbps, 500kbps, CPU 5.7%, MEM 9208KB.
-1.  600 connections,  300Mbps, 500kbps, CPU 18.3%, MEM 13MB.
-1.  900 connections,  450Mbps, 500kbps, CPU 27.9%, MEM 20MB.
-1. 1200 connections,  600Mbps, 500kbps, CPU 43.9%, MEM 26MB.
-1. 1500 connections,  750Mbps, 500kbps, CPU 55.2%, MEM 32MB.
-1. 1800 connections,  900Mbps, 500kbps, CPU 68.8%, MEM 38MB.
-1. 2100 connections, 1050Mbps, 500kbps, CPU 75.7%, MEM 46MB.
-1. 2400 connections, 1200Mbps, 500kbps, CPU 83.7%, MEM 54MB.
-1. 2700 connections, 1350Mbps, 500kbps, CPU 89.9%, MEM 61MB.
+### Publish benchmark
 
-<pre>
-[winlin@dev6 srs]$ dstat
-----total-cpu-usage---- -dsk/total- ---net/lo-- ---paging-- ---system--
-usr sys idl wai hiq siq| read  writ| recv  send|  in   out | int   csw 
- 29  17  39   0   0  15|   0  5325B| 163M  163M|   0     0 |4331  3386 
- 30  16  38   0   0  16|   0  5325B| 160M  160M|   0     0 |4252  3332 
- 30  15  37   0   0  17|   0  7646B| 169M  169M|   0     0 |4015  2886 
- 30  17  36   0   0  17|   0  1638B| 197M  197M|   0     0 |4021  3037 
- 31  17  35   0   0  17|   0   410B| 204M  204M|   0     0 |4181  3243 
- 33  17  32   0   0  18|   0  2185B| 191M  191M|   0     0 |4305  3592 
- 31  15  36   0   0  18|   0  1229B| 127M  127M|   0     0 |4446  3822 
- 34  18  30   0   0  18|   0     0 | 231M  231M|   0     0 |4461  3691 
- 32  17  33   0   0  18|   0   410B| 169M  169M|   0     0 |4518  3788 
-</pre>
+The publish benchmark by st-load:
 
-* See also: [Performance for x86/x64 Test Guide](https://github.com/winlinvip/simple-rtmp-server/wiki/v1_CN_Performance)
-* See also: [Performance for RaspberryPi](https://github.com/winlinvip/simple-rtmp-server/wiki/v1_CN_RaspberryPi)
+* 2014-12-03, SRS 1.0.10,  1k(1000) publishers, xx%CPU, xxMB.
 
 ## Architecture
 

diff --git a/trunk/src/app/srs_app_recv_thread.cpp b/trunk/src/app/srs_app_recv_thread.cpp
@@ -230,11 +230,25 @@ SrsPublishRecvThread::SrsPublishRecvThread(
 
     recv_error_code = ERROR_SUCCESS;
     _nb_msgs = 0;
+    error = st_cond_new();
 }
 
 SrsPublishRecvThread::~SrsPublishRecvThread()
 {
     trd.stop();
+    st_cond_destroy(error);
+}
+
+int SrsPublishRecvThread::wait(int timeout_ms)
+{
+    if (recv_error_code != ERROR_SUCCESS) {
+        return recv_error_code;
+    }
+
+    // ignore any return of cond wait.
+    st_cond_timedwait(error, timeout_ms * 1000);
+
+    return ERROR_SUCCESS;
 }
 
 int64_t SrsPublishRecvThread::nb_msgs()
@@ -282,6 +296,10 @@ int SrsPublishRecvThread::handle(SrsMessage* msg)
 void SrsPublishRecvThread::on_recv_error(int ret)
 {
     recv_error_code = ret;
+
+    // when recv thread error, signal the conn thread to process it.
+    // @see https://github.com/winlinvip/simple-rtmp-server/issues/244
+    st_cond_signal(error);
 }
 
 void SrsPublishRecvThread::on_thread_start()
@@ -294,4 +312,8 @@ void SrsPublishRecvThread::on_thread_stop()
 {
     // we donot set the auto response to true,
     // for we donot set to false yet.
+
+    // when thread stop, signal the conn thread which wait.
+    // @see https://github.com/winlinvip/simple-rtmp-server/issues/244
+    st_cond_signal(error);
 }
diff --git a/trunk/src/app/srs_app_recv_thread.hpp b/trunk/src/app/srs_app_recv_thread.hpp
@@ -142,14 +142,22 @@ class SrsPublishRecvThread : public ISrsMessageHandler
     // the recv thread error code.
     int recv_error_code;
     SrsRtmpConn* _conn;
+    // the params for conn callback.
     SrsSource* _source;
     bool _is_fmle;
     bool _is_edge;
+    // the error timeout cond
+    // @see https://github.com/winlinvip/simple-rtmp-server/issues/244
+    st_cond_t error;
 public:
     SrsPublishRecvThread(SrsRtmpServer* rtmp_sdk, int timeout_ms,
         SrsRtmpConn* conn, SrsSource* source, bool is_fmle, bool is_edge);
     virtual ~SrsPublishRecvThread();
 public:
+    /**
+    * wait for error for some timeout.
+    */
+    virtual int wait(int timeout_ms);
     virtual int64_t nb_msgs();
     virtual int error_code();
 public:

diff --git a/trunk/src/app/srs_app_rtmp_conn.cpp b/trunk/src/app/srs_app_rtmp_conn.cpp
@@ -750,17 +750,15 @@ int SrsRtmpConn::do_publishing(SrsSource* source, SrsPublishRecvThread* trd)
 
     int64_t nb_msgs = 0;
     while (true) {
-        // use small loop to check the error code, interval = 30s/43 = 697ms.
-        for (int i = 0; i < 43; i++) {
-            st_usleep(SRS_CONSTS_RTMP_RECV_TIMEOUT_US / 43);
+        // cond wait for error.
+        trd->wait(SRS_CONSTS_RTMP_RECV_TIMEOUT_US / 1000);
 
-            // check the thread error code.
-            if ((ret = trd->error_code()) != ERROR_SUCCESS) {
-                if (!srs_is_client_gracefully_close(ret)) {
-                    srs_error("recv thread failed. ret=%d", ret);
-                }
-                return ret;
+        // check the thread error code.
+        if ((ret = trd->error_code()) != ERROR_SUCCESS) {
+            if (!srs_is_client_gracefully_close(ret)) {
+                srs_error("recv thread failed. ret=%d", ret);
             }
+            return ret;
         }
 
         // when not got any messages, timeout.