API: support metrics for prometheus.(#2899) (#3189)

* API: support metrics for prometheus.

* Metrics: optimize metrics statistics info.

* Refine: remove redundant code.

* Refine: fix metrics srs_streams param.

* Metrics: add major param.

* Metrics: refine params and metric comments.

* For #2899: API: Support exporter for Prometheus. v5.0.67

Co-authored-by: winlin <winlin@vip.126.com>

trunk/conf/full.conf

@@ -657,6 +657,21 @@ tencentcloud_apm {
     debug_logging off;
 }
 
+# Prometheus exporter config.
+# See https://prometheus.io/docs/instrumenting/exporters
+exporter {
+    # Whether exporter is enabled.
+    # Overwrite by env SRS_EXPORTER_ENABLED
+    # Default: off
+    enabled         off;
+    # The logging label to category the cluster servers.
+    # Overwrite by env SRS_EXPORTER_LABEL
+    label cn-beijing;
+    # The logging tag to category the cluster servers.
+    # Overwrite by env SRS_EXPORTER_TAG
+    tag cn-edge;
+}
+
 #############################################################################################
 # heartbeat/stats sections
 #############################################################################################

trunk/doc/CHANGELOG.md

@@ -7,6 +7,7 @@ The changelog for SRS.
 
 ## SRS 5.0 Changelog
 
+* v5.0, 2022-09-27, For [#2899](https://github.com/ossrs/srs/issues/2899): API: Support exporter for Prometheus. v5.0.67
 * v5.0, 2022-09-27, For [#3167](https://github.com/ossrs/srs/issues/3167): WebRTC: Refine sequence jitter algorithm. v5.0.66
 * v5.0, 2022-09-22, Fix [#3164](https://github.com/ossrs/srs/issues/3164): SRT: Choppy when audio ts gap is too large. v5.0.65
 * v5.0, 2022-09-16, APM: Support distributed tracing by Tencent Cloud APM. v5.0.64

trunk/src/app/srs_app_config.cpp

@@ -2227,6 +2227,7 @@ srs_error_t SrsConfig::check_normal_config()
             && n != "inotify_auto_reload" && n != "auto_reload_for_docker" && n != "tcmalloc_release_rate"
             && n != "query_latest_version" && n != "first_wait_for_qlv" && n != "threads"
             && n != "circuit_breaker" && n != "is_full" && n != "in_docker" && n != "tencentcloud_cls"
+            && n != "exporter"
             ) {
             return srs_error_new(ERROR_SYSTEM_CONFIG_INVALID, "illegal directive %s", n.c_str());
         }
@@ -3563,6 +3564,63 @@ bool SrsConfig::get_tencentcloud_apm_debug_logging()
     return SRS_CONF_PERFER_FALSE(conf->arg0());
 }
 
+bool SrsConfig::get_exporter_enabled()
+{
+    SRS_OVERWRITE_BY_ENV_BOOL("SRS_EXPORTER_ENABLED");
+
+    static bool DEFAULT = false;
+
+    SrsConfDirective* conf = root->get("exporter");
+    if (!conf) {
+        return DEFAULT;
+    }
+
+    conf = conf->get("enabled");
+    if (!conf) {
+        return DEFAULT;
+    }
+
+    return SRS_CONF_PERFER_FALSE(conf->arg0());
+}
+
+string SrsConfig::get_exporter_label()
+{
+    SRS_OVERWRITE_BY_ENV_STRING("SRS_EXPORTER_LABEL");
+
+    static string DEFAULT = "";
+
+    SrsConfDirective* conf = root->get("exporter");
+    if (!conf) {
+        return DEFAULT;
+    }
+
+    conf = conf->get("label");
+    if (!conf) {
+        return DEFAULT;
+    }
+
+    return conf->arg0();
+}
+
+string SrsConfig::get_exporter_tag()
+{
+    SRS_OVERWRITE_BY_ENV_STRING("SRS_EXPORTER_TAG");
+
+    static string DEFAULT = "";
+
+    SrsConfDirective* conf = root->get("exporter");
+    if (!conf) {
+        return DEFAULT;
+    }
+
+    conf = conf->get("tag");
+    if (!conf) {
+        return DEFAULT;
+    }
+
+    return conf->arg0();
+}
+
 vector<SrsConfDirective*> SrsConfig::get_stream_casters()
 {
     srs_assert(root);

trunk/src/app/srs_app_config.hpp

@@ -1086,6 +1086,11 @@ public:
     // The device name configed in args of directive.
     // @return the disk device name to stat. NULL if not configed.
     virtual SrsConfDirective* get_stats_disk_device();
+public:
+    // Get Prometheus exporter config.
+    virtual bool get_exporter_enabled();
+    virtual std::string get_exporter_label();
+    virtual std::string get_exporter_tag();
 };
 
 #endif

trunk/src/app/srs_app_http_api.cpp

@@ -1062,3 +1062,96 @@ srs_error_t SrsGoApiTcmalloc::serve_http(ISrsHttpResponseWriter* w, ISrsHttpMess
 }
 #endif
 
+
+SrsGoApiMetrics::SrsGoApiMetrics()
+{
+    enabled_ = _srs_config->get_exporter_enabled();
+    label_ = _srs_config->get_exporter_label();
+    tag_ = _srs_config->get_exporter_tag();
+}
+
+SrsGoApiMetrics::~SrsGoApiMetrics()
+{
+}
+
+srs_error_t SrsGoApiMetrics::serve_http(ISrsHttpResponseWriter* w, ISrsHttpMessage* r)
+{
+    // whether enabled the HTTP Metrics API.
+    if (!enabled_) {
+        return srs_api_response_code(w, r, ERROR_EXPORTER_DISABLED);
+    }
+
+    /*
+     * build_info gauge
+     * send_bytes_total counter
+     * receive_bytes_total counter
+     * streams gauge
+     * clients gauge
+     * clients_total counter
+     * error counter
+    */
+
+    SrsStatistic* stat = SrsStatistic::instance();
+
+    std::stringstream ss;
+
+    // Build info from Config.
+    ss << "# HELP srs_build_info A metric with a constant '1' value labeled by build_date, version from which SRS was built.\n"
+        << "# TYPE srs_build_info gauge\n"
+        << "srs_build_info{"
+            << "build_date=\"" << SRS_BUILD_DATE << "\","
+            << "major=\"" << VERSION_MAJOR << "\","
+            << "version=\"" << RTMP_SIG_SRS_VERSION << "\","
+            << "code=\"" << RTMP_SIG_SRS_CODE<< "\"";
+    if (!label_.empty()) ss << ",label=\"" << label_ << "\"";
+    if (!tag_.empty()) ss << ",tag=\"" << tag_ << "\"";
+    ss << "} 1\n";
+
+    // Dump metrics by statistic.
+    int64_t send_bytes, recv_bytes, nstreams, nclients, total_nclients, nerrs;
+    stat->dumps_metrics(send_bytes, recv_bytes, nstreams, nclients, total_nclients, nerrs);
+
+    // The total of bytes sent.
+    ss << "# HELP srs_send_bytes_total SRS total sent bytes.\n"
+       << "# TYPE srs_send_bytes_total counter\n"
+       << "srs_send_bytes_total "
+       << send_bytes
+       << "\n";
+
+    // The total of bytes received.
+    ss << "# HELP srs_receive_bytes_total SRS total received bytes.\n"
+       << "# TYPE srs_receive_bytes_total counter\n"
+       << "srs_receive_bytes_total "
+       << recv_bytes
+       << "\n";
+
+    // Current number of online streams.
+    ss << "# HELP srs_streams The number of SRS concurrent streams.\n"
+       << "# TYPE srs_streams gauge\n"
+       << "srs_streams "
+       << nstreams
+       << "\n";
+
+    // Current number of online clients.
+    ss << "# HELP srs_clients The number of SRS concurrent clients.\n"
+       << "# TYPE srs_clients gauge\n"
+       << "srs_clients "
+       << nclients
+       << "\n";
+
+    // The total of clients connections.
+    ss << "# HELP srs_clients_total The total counts of SRS clients.\n"
+       << "# TYPE srs_clients_total counter\n"
+       << "srs_clients_total "
+       << total_nclients
+       << "\n";
+
+    // The total of clients errors.
+    ss << "# HELP srs_clients_errs_total The total errors of SRS clients.\n"
+       << "# TYPE srs_clients_errs_total counter\n"
+       << "srs_clients_errs_total "
+       << nerrs
+       << "\n";
+
+    return srs_api_response(w, r, ss.str());
+}

trunk/src/app/srs_app_http_api.hpp

@@ -216,5 +216,18 @@ public:
 };
 #endif
 
+class SrsGoApiMetrics : public ISrsHttpHandler
+{
+private:
+    bool enabled_;
+    std::string label_;
+    std::string tag_;
+public:
+    SrsGoApiMetrics();
+    virtual ~SrsGoApiMetrics();
+public:
+    virtual srs_error_t serve_http(ISrsHttpResponseWriter* w, ISrsHttpMessage* r);
+};
+
 #endif
 

trunk/src/app/srs_app_http_conn.cpp

@@ -406,7 +406,7 @@ srs_error_t SrsHttpxConn::on_conn_done(srs_error_t r0)
 {
     // Only stat the HTTP streaming clients, ignore all API clients.
     if (enable_stat_) {
-        SrsStatistic::instance()->on_disconnect(get_id().c_str());
+        SrsStatistic::instance()->on_disconnect(get_id().c_str(), r0);
         SrsStatistic::instance()->kbps_add_delta(get_id().c_str(), conn->delta());
     }
 

trunk/src/app/srs_app_http_static.cpp

@@ -331,7 +331,8 @@ srs_error_t SrsHlsStream::on_timer(srs_utime_t interval)
             http_hooks_on_stop(info->req);
 
             SrsStatistic* stat = SrsStatistic::instance();
-            stat->on_disconnect(ctx);
+            // TODO: FIXME: Should finger out the err.
+            stat->on_disconnect(ctx, srs_success);
 
             map_ctx_info_.erase(it);
             srs_freep(info);

trunk/src/app/srs_app_rtc_conn.cpp

@@ -458,7 +458,8 @@ SrsRtcPlayStream::~SrsRtcPlayStream()
 	
     // update the statistic when client coveried.
     SrsStatistic* stat = SrsStatistic::instance();
-    stat->on_disconnect(cid_.c_str());
+    // TODO: FIXME: Should finger out the err.
+    stat->on_disconnect(cid_.c_str(), srs_success);
 }
 
 srs_error_t SrsRtcPlayStream::initialize(SrsRequest* req, std::map<uint32_t, SrsRtcTrackDescription*> sub_relations)
@@ -1108,7 +1109,8 @@ SrsRtcPublishStream::~SrsRtcPublishStream()
 	
     // update the statistic when client coveried.
     SrsStatistic* stat = SrsStatistic::instance();
-    stat->on_disconnect(cid_.c_str());
+    // TODO: FIXME: Should finger out the err.
+    stat->on_disconnect(cid_.c_str(), srs_success);
 }
 
 srs_error_t SrsRtcPublishStream::initialize(SrsRequest* r, SrsRtcSourceDescription* stream_desc)

trunk/src/app/srs_app_rtc_network.cpp

@@ -749,7 +749,7 @@ srs_error_t SrsRtcTcpConn::cycle()
     srs_error_t err = do_cycle();
 
     // Only stat the HTTP streaming clients, ignore all API clients.
-    SrsStatistic::instance()->on_disconnect(get_id().c_str());
+    SrsStatistic::instance()->on_disconnect(get_id().c_str(), err);
     SrsStatistic::instance()->kbps_add_delta(get_id().c_str(), delta_);
 
     // Because we use manager to manage this object, not the http connection object, so we must remove it here.

trunk/src/app/srs_app_rtmp_conn.cpp

@@ -1551,7 +1551,7 @@ srs_error_t SrsRtmpConn::cycle()
     // Update statistic when done.
     SrsStatistic* stat = SrsStatistic::instance();
     stat->kbps_add_delta(get_id().c_str(), delta_);
-    stat->on_disconnect(get_id().c_str());
+    stat->on_disconnect(get_id().c_str(), err);
 
     // Notify manager to remove it.
     // Note that we create this object, so we use manager to remove it.

trunk/src/app/srs_app_server.cpp

@@ -887,6 +887,10 @@ srs_error_t SrsServer::http_handle()
         return srs_error_wrap(err, "handle tests errors");
     }
 #endif
+    // metrics by prometheus
+    if ((err = http_api_mux->handle("/metrics", new SrsGoApiMetrics())) != srs_success) {
+        return srs_error_wrap(err, "handle tests errors");
+    }
     
     // TODO: FIXME: for console.
     // TODO: FIXME: support reload.

trunk/src/app/srs_app_srt_conn.cpp

@@ -228,7 +228,7 @@ srs_error_t SrsMpegtsSrtConn::cycle()
     // Update statistic when done.
     SrsStatistic* stat = SrsStatistic::instance();
     stat->kbps_add_delta(get_id().c_str(), delta_);
-    stat->on_disconnect(get_id().c_str());
+    stat->on_disconnect(get_id().c_str(), err);
 
     // Notify manager to remove it.
     // Note that we create this object, so we use manager to remove it.

trunk/src/app/srs_app_statistic.cpp

@@ -241,6 +241,9 @@ SrsStatistic* SrsStatistic::_instance = NULL;
 SrsStatistic::SrsStatistic()
 {
     kbps = new SrsKbps();
+
+    nb_clients_ = 0;
+    nb_errs_ = 0;
 }
 
 SrsStatistic::~SrsStatistic()
@@ -421,11 +424,13 @@ srs_error_t SrsStatistic::on_client(std::string id, SrsRequest* req, ISrsExpire*
     // @see https://github.com/ossrs/srs/issues/2311
     srs_freep(client->req);
     client->req = req->copy();
+
+    nb_clients_++;
     
     return err;
 }
 
-void SrsStatistic::on_disconnect(std::string id)
+void SrsStatistic::on_disconnect(std::string id, srs_error_t err)
 {
     std::map<std::string, SrsStatisticClient*>::iterator it = clients.find(id);
     if (it == clients.end()) return;
@@ -440,6 +445,10 @@ void SrsStatistic::on_disconnect(std::string id)
     stream->nb_clients--;
     vhost->nb_clients--;
 
+    if (srs_error_code(err) != ERROR_SUCCESS) {
+        nb_errs_++;
+    }
+
     cleanup_stream(stream);
 }
 
@@ -721,3 +730,19 @@ SrsStatisticStream* SrsStatistic::create_stream(SrsStatisticVhost* vhost, SrsReq
     return stream;
 }
 
+srs_error_t SrsStatistic::dumps_metrics(int64_t& send_bytes, int64_t& recv_bytes, int64_t& nstreams, int64_t& nclients, int64_t& total_nclients, int64_t& nerrs)
+{
+    srs_error_t err = srs_success;
+
+    send_bytes = kbps->get_send_bytes();
+    recv_bytes = kbps->get_recv_bytes();
+
+    nstreams = streams.size();
+    nclients = clients.size();
+
+    total_nclients = nb_clients_;
+    nerrs = nb_errs_;
+
+    return err;
+}
+

trunk/src/app/srs_app_statistic.hpp

@@ -139,6 +139,11 @@ private:
     std::map<std::string, SrsStatisticClient*> clients;
     // The server total kbps.
     SrsKbps* kbps;
+private:
+    // The total of clients connections.
+    int64_t nb_clients_;
+    // The total of clients errors.
+    int64_t nb_errs_;
 private:
     SrsStatistic();
     virtual ~SrsStatistic();
@@ -177,7 +182,7 @@ public:
     // @remark the on_disconnect always call, while the on_client is call when
     //      only got the request object, so the client specified by id maybe not
     //      exists in stat.
-    virtual void on_disconnect(std::string id);
+    virtual void on_disconnect(std::string id, srs_error_t err);
 private:
     // Cleanup the stream if stream is not active and for the last client.
     void cleanup_stream(SrsStatisticStream* stream);
@@ -210,6 +215,9 @@ public:
 private:
     virtual SrsStatisticVhost* create_vhost(SrsRequest* req);
     virtual SrsStatisticStream* create_stream(SrsStatisticVhost* vhost, SrsRequest* req);
+public:
+    // Dumps exporter metrics.
+    virtual srs_error_t dumps_metrics(int64_t& send_bytes, int64_t& recv_bytes, int64_t& nstreams, int64_t& nclients, int64_t& total_nclients, int64_t& nerrs);
 };
 
 // Generate a random string id, with constant prefix.

trunk/src/core/srs_core_version5.hpp

@@ -9,6 +9,6 @@
 
 #define VERSION_MAJOR       5
 #define VERSION_MINOR       0
-#define VERSION_REVISION    66
+#define VERSION_REVISION    67
 
 #endif

trunk/src/kernel/srs_kernel_error.hpp

@@ -100,6 +100,7 @@
     XX(ERROR_APM_EXCEED_SIZE               , 1087, "ApmExceedSize", "APM logs exceed max size 5MB") \
     XX(ERROR_APM_ENDPOINT                  , 1088, "ApmEndpoint", "APM endpoint is invalid") \
     XX(ERROR_APM_AUTH                      , 1089, "ApmAuth", "APM team or token is invalid") \
+    XX(ERROR_EXPORTER_DISABLED             , 1090, "ExporterDisable", "Prometheus exporter is disabled") \
 
 /**************************************************/
 /* RTMP protocol error. */