diff --git a/python/ray/includes/ray_config.pxd b/python/ray/includes/ray_config.pxd
index 791ff9059..17264b854 100644
--- a/python/ray/includes/ray_config.pxd
+++ b/python/ray/includes/ray_config.pxd
@@ -90,3 +90,5 @@ cdef extern from "ray/common/ray_config.h" nogil:
         c_bool put_small_object_in_memory_store() const
 
         uint32_t max_tasks_in_flight_per_worker() const
+
+        uint64_t metrics_report_interval_ms() const
diff --git a/python/ray/includes/ray_config.pxi b/python/ray/includes/ray_config.pxi
index 725140389..3dd256010 100644
--- a/python/ray/includes/ray_config.pxi
+++ b/python/ray/includes/ray_config.pxi
@@ -165,3 +165,7 @@ cdef class Config:
     @staticmethod
     def max_tasks_in_flight_per_worker():
         return RayConfig.instance().max_tasks_in_flight_per_worker()
+
+    @staticmethod
+    def metrics_report_interval_ms():
+        return RayConfig.instance().metrics_report_interval_ms()
diff --git a/python/ray/tests/test_metrics_agent.py b/python/ray/tests/test_metrics_agent.py
index f88f36b26..c9254ce5a 100644
--- a/python/ray/tests/test_metrics_agent.py
+++ b/python/ray/tests/test_metrics_agent.py
@@ -14,6 +14,7 @@ from ray.core.generated.common_pb2 import MetricPoint
 from ray.dashboard.util import get_unused_port
 from ray.metrics_agent import (Gauge, MetricsAgent,
                                PrometheusServiceDiscoveryWriter)
+from ray.test_utils import wait_for_condition
 
 
 def generate_metrics_point(name: str,
@@ -216,6 +217,80 @@ def test_prometheus_file_based_service_discovery(ray_start_cluster):
         loaded_json_data["targets"]))
 
 
+def test_metrics_export_end_to_end(ray_start_cluster):
+    NUM_NODES = 2
+    cluster = ray_start_cluster
+    # Add a head node.
+    cluster.add_node(
+        _internal_config=json.dumps({
+            "metrics_report_interval_ms": 1000
+        }))
+    # Add worker nodes.
+    [cluster.add_node() for _ in range(NUM_NODES - 1)]
+    cluster.wait_for_nodes()
+    ray.init(address=cluster.address)
+
+    # Generate some metrics around actor & tasks.
+    @ray.remote
+    def f():
+        return 3
+
+    @ray.remote
+    class A:
+        def ping(self):
+            return 3
+
+    ray.get([f.remote() for _ in range(30)])
+    a = A.remote()
+    ray.get(a.ping.remote())
+
+    node_info_list = ray.nodes()
+    prom_addresses = []
+    for node_info in node_info_list:
+        metrics_export_port = node_info["MetricsExportPort"]
+        addr = node_info["NodeManagerAddress"]
+        prom_addresses.append(f"{addr}:{metrics_export_port}")
+
+    # Make sure we can ping Prometheus endpoints.
+    def get_component_information(prom_addresses):
+        # TODO(sang): Add a core worker & gcs_server after adding metrics.
+        components_dict = {}
+        for address in prom_addresses:
+            if address not in components_dict:
+                components_dict[address] = set()
+            try:
+                response = requests.get(
+                    "http://localhost:{}".format(metrics_export_port))
+            except requests.exceptions.ConnectionError:
+                return components_dict
+
+            for line in response.text.split("\n"):
+                for family in text_string_to_metric_families(line):
+                    for sample in family.samples:
+                        # print(sample)
+                        if "Component" in sample.labels:
+                            components_dict[address].add(
+                                sample.labels["Component"])
+        return components_dict
+
+    def test_prometheus_endpoint():
+        # TODO(sang): Add a core worker & gcs_server after adding metrics.
+        components_dict = get_component_information(prom_addresses)
+        COMPONENTS_CANDIDATES = {"raylet"}
+        return all(
+            COMPONENTS_CANDIDATES.issubset(components)
+            for components in components_dict.values())
+
+    try:
+        wait_for_condition(test_prometheus_endpoint, timeout=3)
+    except RuntimeError:
+        # This is for debugging when test failed.
+        print(get_component_information(prom_addresses))
+        raise RuntimeError("All components were not visible to "
+                           "prometheus endpoints on time.")
+    ray.shutdown()
+
+
 if __name__ == "__main__":
     import sys
     sys.exit(pytest.main(["-v", __file__]))
diff --git a/src/ray/common/ray_config_def.h b/src/ray/common/ray_config_def.h
index f3607caa5..ae7cdd502 100644
--- a/src/ray/common/ray_config_def.h
+++ b/src/ray/common/ray_config_def.h
@@ -346,3 +346,6 @@ RAY_CONFIG(bool, enable_multi_tenancy, false)
 
 /// Whether start the Plasma Store as a Raylet thread.
 RAY_CONFIG(bool, ownership_based_object_directory_enabled, false)
+
+// The interval where metrics are exported in milliseconds.
+RAY_CONFIG(uint64_t, metrics_report_interval_ms, 10000)
diff --git a/src/ray/gcs/gcs_server/gcs_server_main.cc b/src/ray/gcs/gcs_server/gcs_server_main.cc
index a7eaba2b6..e4ee94fad 100644
--- a/src/ray/gcs/gcs_server/gcs_server_main.cc
+++ b/src/ray/gcs/gcs_server/gcs_server_main.cc
@@ -57,7 +57,7 @@ int main(int argc, char *argv[]) {
   }
 
   RayConfig::instance().initialize(config_map);
-  const ray::stats::TagsType global_tags = {{ray::stats::JobNameKey, "gcs_server"},
+  const ray::stats::TagsType global_tags = {{ray::stats::ComponentKey, "gcs_server"},
                                             {ray::stats::VersionKey, "0.9.0.dev0"}};
   ray::stats::Init(global_tags, metrics_agent_port);
 
diff --git a/src/ray/stats/metric.h b/src/ray/stats/metric.h
index 4a5b7ff70..06e8534c4 100644
--- a/src/ray/stats/metric.h
+++ b/src/ray/stats/metric.h
@@ -76,12 +76,12 @@ class StatsConfig final {
   /// If true, don't collect metrics in this process.
   bool is_stats_disabled_ = true;
   // Regular reporting interval for all reporters.
-  absl::Duration report_interval_ = absl::Seconds(10);
+  absl::Duration report_interval_ = absl::Milliseconds(10000);
   // Time interval for periodic aggregation.
   // Exporter may capture empty collection if harvest interval is longer than
   // report interval. So harvest interval is suggusted to be half of report
   // interval.
-  absl::Duration harvest_interval_ = absl::Seconds(5);
+  absl::Duration harvest_interval_ = absl::Milliseconds(5000);
   // Whether or not if the stats has been initialized.
   bool is_initialized_ = false;
 };
diff --git a/src/ray/stats/stats.h b/src/ray/stats/stats.h
index 2f85f2549..c7bb283e0 100644
--- a/src/ray/stats/stats.h
+++ b/src/ray/stats/stats.h
@@ -89,6 +89,13 @@ static inline void Init(const TagsType &global_tags, const int metrics_agent_por
     exporter = exporter_to_use;
   }
 
+  // Set interval.
+  StatsConfig::instance().SetReportInterval(absl::Milliseconds(std::max(
+      RayConfig::instance().metrics_report_interval_ms(), static_cast<uint64_t>(1000))));
+  StatsConfig::instance().SetHarvestInterval(
+      absl::Milliseconds(std::max(RayConfig::instance().metrics_report_interval_ms() / 2,
+                                  static_cast<uint64_t>(500))));
+
   MetricExporter::Register(exporter, metrics_report_batch_size);
   opencensus::stats::StatsExporter::SetInterval(
       StatsConfig::instance().GetReportInterval());
@@ -101,7 +108,6 @@ static inline void Init(const TagsType &global_tags, const int metrics_agent_por
 /// Shutdown the initialized stats library.
 /// This cleans up various threads and metadata for stats library.
 static inline void Shutdown() {
-  // TODO(sang): Harvest thread is not currently cleaned up.
   absl::MutexLock lock(&stats_mutex);
   if (!StatsConfig::instance().IsInitialized()) {
     // Return if stats had never been initialized.