Set http address for metrics when http service is disable (triton-inf…

…erence-server#4175)

* Set address for metrics when http service is disable

* Update comment

* Update documentation

docs/metrics.md

@@ -42,7 +42,10 @@ $ curl localhost:8002/metrics
 The tritonserver --allow-metrics=false option can be used to disable
 all metric reporting and --allow-gpu-metrics=false can be used to
 disable just the GPU Utilization and GPU Memory metrics. The
---metrics-port option can be used to select a different port.
+--metrics-port option can be used to select a different port. For now,
+Triton reuses http address for metrics endpoint. The option --http-address
+can be used to bind http and metrics endpoints to the same specific address
+when http service is enabled.
 
 The following table describes the available metrics.
 

src/main.cc

@@ -133,6 +133,11 @@ std::unique_ptr<triton::server::HTTPServer> metrics_service_;
 bool allow_metrics_ = true;
 int32_t metrics_port_ = 8002;
 float metrics_interval_ms_ = 2000;
+#ifndef TRITON_ENABLE_HTTP
+// Triton uses the same address for http and metrics services.
+// Need to set http address for metrics when http service is disable.
+std::string http_address_ = "0.0.0.0";
+#endif  // NOT TRITON_ENABLE_HTTP
 #endif  // TRITON_ENABLE_METRICS
 
 #ifdef TRITON_ENABLE_TRACING