Merge pull request #804 from zyearn/customize_brpc_metrics_path

customize brpc metrics path

docs/cn/bvar.md

@@ -92,4 +92,4 @@ process_username : "gejun"
 
 # bvar导出到其它监控系统格式
 
-bvar已支持的其它监控系统格式有[Prometheus](https://prometheus.io)。将Prometheus的抓取url地址的路径设置为`/metrics`即可，例如brpc server跑在本机的8080端口，则抓取url配置为`127.0.0.1:8080/metrics`。
+bvar已支持的其它监控系统格式有[Prometheus](https://prometheus.io)。将Prometheus的抓取url地址的路径设置为`/brpc_metrics`即可，例如brpc server跑在本机的8080端口，则抓取url配置为`127.0.0.1:8080/brpc_metrics`。

docs/en/bvar.md

@@ -92,4 +92,4 @@ The monitoring system should combine data on every single machine periodically a
 
 # Dump to the format of other monitoring system
 
-Currently monitoring system supported by bvar is [Prometheus](https://prometheus.io). All you need to do is to set the path in scraping target url to `/metrics`. For example, if brpc server is running in localhost on port 8080, the scraping target should be `127.0.0.1:8080/metrics`.
+Currently monitoring system supported by bvar is [Prometheus](https://prometheus.io). All you need to do is to set the path in scraping target url to `/brpc_metrics`. For example, if brpc server is running on localhost:8080, the scraping target should be `127.0.0.1:8080/brpc_metrics`.

src/brpc/builtin/prometheus_metrics_service.cpp

@@ -32,6 +32,9 @@ DECLARE_int32(bvar_latency_p3);
 
 namespace brpc {
 
+// Defined in server.cpp
+extern const char* const g_server_info_prefix;
+
 // This is a class that convert bvar result to prometheus output.
 // Currently the output only includes gauge and summary for two
 // reasons:
@@ -61,8 +64,8 @@ private:
 
     struct SummaryItems {
         std::string latency_percentiles[NPERCENTILES];
-        std::string latency_avg;
-        std::string count;
+        int64_t latency_avg;
+        int64_t count;
         std::string metric_name;
 
         bool IsComplete() const { return !metric_name.empty(); }
@@ -103,6 +106,7 @@ PrometheusMetricsDumper::ProcessLatencyRecorderSuffix(const butil::StringPiece&
         "_latency_999", "_latency_9999", "_max_latency"
     };
     CHECK(NPERCENTILES == arraysize(latency_names));
+    const std::string desc_str = desc.as_string();
     butil::StringPiece metric_name(name);
     for (int i = 0; i < NPERCENTILES; ++i) {
         if (!metric_name.ends_with(latency_names[i])) {
@@ -110,7 +114,7 @@ PrometheusMetricsDumper::ProcessLatencyRecorderSuffix(const butil::StringPiece&
         }
         metric_name.remove_suffix(latency_names[i].size());
         SummaryItems* si = &_m[metric_name.as_string()];
-        si->latency_percentiles[i] = desc.as_string();
+        si->latency_percentiles[i] = desc_str;
         if (i == NPERCENTILES - 1) {
             // '_max_latency' is the last suffix name that appear in the sorted bvar
             // list, which means all related percentiles have been gathered and we are
@@ -123,13 +127,13 @@ PrometheusMetricsDumper::ProcessLatencyRecorderSuffix(const butil::StringPiece&
     if (metric_name.ends_with("_latency")) {
         metric_name.remove_suffix(8);
         SummaryItems* si = &_m[metric_name.as_string()];
-        si->latency_avg = desc.as_string();
+        si->latency_avg = strtoll(desc_str.data(), NULL, 10);
         return si;
     }
     if (metric_name.ends_with("_count")) {
         metric_name.remove_suffix(6);
         SummaryItems* si = &_m[metric_name.as_string()];
-        si->count = desc.as_string();
+        si->count = strtoll(desc_str.data(), NULL, 10);
         return si;
     }
     return NULL;
@@ -168,8 +172,7 @@ bool PrometheusMetricsDumper::DumpLatencyRecorderSuffix(
          << si->metric_name << "_sum "
          // There is no sum of latency in bvar output, just use
          // average * count as approximation
-         << strtoll(si->latency_avg.data(), NULL, 10) *
-                strtoll(si->count.data(), NULL, 10) << '\n'
+         << si->latency_avg * si->count << '\n'
          << si->metric_name << "_count " << si->count << '\n';
     return true;
 }
@@ -181,14 +184,21 @@ void PrometheusMetricsService::default_method(::google::protobuf::RpcController*
     ClosureGuard done_guard(done);
     Controller *cntl = static_cast<Controller*>(cntl_base);
     cntl->http_response().set_content_type("text/plain");
+    if (DumpPrometheusMetricsToIOBuf(&cntl->response_attachment()) != 0) {
+        cntl->SetFailed("Fail to dump metrics");
+        return;
+    }
+}
+
+int DumpPrometheusMetricsToIOBuf(butil::IOBuf* output) {
     butil::IOBufBuilder os;
-    PrometheusMetricsDumper dumper(&os, _server->ServerPrefix());
+    PrometheusMetricsDumper dumper(&os, g_server_info_prefix);
     const int ndump = bvar::Variable::dump_exposed(&dumper, NULL);
     if (ndump < 0) {
-        cntl->SetFailed("Fail to dump metrics");
-        return;
+        return -1;
     }
-    os.move_to(cntl->response_attachment());
+    os.move_to(*output);
+    return 0;
 }
 
 } // namespace brpc

src/brpc/builtin/prometheus_metrics_service.h

@@ -18,23 +18,19 @@
 #define BRPC_PROMETHEUS_METRICS_SERVICE_H
 
 #include "brpc/builtin_service.pb.h"
-#include "brpc/server.h"
 
 namespace brpc {
 
-class PrometheusMetricsService : public metrics {
+class PrometheusMetricsService : public brpc_metrics {
 public:
-    PrometheusMetricsService(Server* server)
-        : _server(server) {}
-
     void default_method(::google::protobuf::RpcController* cntl_base,
                         const ::brpc::MetricsRequest* request,
                         ::brpc::MetricsResponse* response,
                         ::google::protobuf::Closure* done) override;
-private:
-    Server* _server;
 };
 
+int DumpPrometheusMetricsToIOBuf(butil::IOBuf* output);
+
 } // namepace brpc
 
 #endif  // BRPC_PROMETHEUS_METRICS_SERVICE_H

src/brpc/builtin_service.proto

@@ -97,7 +97,7 @@ service sockets {
     rpc default_method(SocketsRequest) returns (SocketsResponse);
 }
 
-service metrics {
+service brpc_metrics {
     rpc default_method(MetricsRequest) returns (MetricsResponse);
 }
 

src/brpc/details/server_private_accessor.h

@@ -22,7 +22,6 @@
 #include "brpc/builtin/bad_method_service.h"
 #include "brpc/restful.h"
 
-
 namespace brpc {
 
 // A wrapper to access some private methods/fields of `Server'

src/brpc/server.cpp

@@ -92,6 +92,8 @@ namespace brpc {
 BAIDU_CASSERT(sizeof(int32_t) == sizeof(butil::subtle::Atomic32),
               Atomic32_must_be_int32);
 
+extern const char* const g_server_info_prefix = "rpc_server";
+
 const char* status_str(Server::Status s) {
     switch (s) {
     case Server::UNINITIALIZED: return "UNINITIALIZED";
@@ -266,7 +268,7 @@ static bvar::Vector<unsigned, 2> GetSessionLocalDataCount(void* arg) {
 }
 
 std::string Server::ServerPrefix() const {
-    return butil::string_printf("rpc_server_%d", listen_address().port);
+    return butil::string_printf("%s_%d", g_server_info_prefix, listen_address().port);
 }
 
 void* Server::UpdateDerivedVars(void* arg) {
@@ -484,7 +486,7 @@ int Server::AddBuiltinServices() {
         LOG(ERROR) << "Fail to add ListService";
         return -1;
     }
-    if (AddBuiltinService(new (std::nothrow) PrometheusMetricsService(this))) {
+    if (AddBuiltinService(new (std::nothrow) PrometheusMetricsService)) {
         LOG(ERROR) << "Fail to add MetricsService";
         return -1;
     }

test/brpc_prometheus_metrics_unittest.cpp

@@ -40,12 +40,17 @@ TEST(PrometheusMetrics, sanity) {
     ASSERT_EQ(0, server.AddService(&echo_svc, brpc::SERVER_DOESNT_OWN_SERVICE));
     ASSERT_EQ(0, server.Start("127.0.0.1:8614", NULL));
 
+    brpc::Server server2;
+    DummyEchoServiceImpl echo_svc2;
+    ASSERT_EQ(0, server2.AddService(&echo_svc2, brpc::SERVER_DOESNT_OWN_SERVICE));
+    ASSERT_EQ(0, server2.Start("127.0.0.1:8615", NULL));
+
     brpc::Channel channel;
     brpc::ChannelOptions channel_opts;
     channel_opts.protocol = "http";
     ASSERT_EQ(0, channel.Init("127.0.0.1:8614", &channel_opts));
     brpc::Controller cntl;
-    cntl.http_request().uri() = "/metrics";
+    cntl.http_request().uri() = "/brpc_metrics";
     channel.CallMethod(NULL, &cntl, NULL, NULL, NULL);
     ASSERT_FALSE(cntl.Failed());
     std::string res = cntl.response_attachment().to_string();