updated queries

robusta-dev · Jul 23, 2024 · e302a5c · e302a5c
1 parent 360acce
commit e302a5c
Show file tree

Hide file tree

Showing 2 changed files with 7 additions and 7 deletions.
diff --git a/src/robusta/core/sinks/robusta/dal/supabase_dal.py b/src/robusta/core/sinks/robusta/dal/supabase_dal.py
@@ -556,13 +556,13 @@ def publish_namespaces(self, namespaces: List[NamespaceInfo]):
             logging.error(f"Failed to persist namespaces {namespaces} error: {e}")
             raise
 
-    def publish_cluster_nodes(self, node_count: int, pod_count: int, avg_cpu: float, avg_mem: float):
+    def publish_cluster_nodes(self, node_count: int, pod_count: int, avg_cpu: Optional[float] = None, avg_mem: Optional[float] = None):
         data = {
             "_account_id": self.account_id,
             "_cluster_id": self.cluster,
             "_node_count": node_count,
             "_cpu_utilization": avg_cpu,
-            "_memory_utilization": int(avg_mem),
+            "_memory_utilization": avg_mem,
             "_pod_count": pod_count
         }
         try:

diff --git a/src/robusta/core/sinks/robusta/prometheus_discovery_utils.py b/src/robusta/core/sinks/robusta/prometheus_discovery_utils.py
@@ -40,19 +40,19 @@ def get_status(self) -> PrometheusHealthStatus:
 
     def get_cluster_avg_cpu(self) -> Optional[float]:
         cpu_query = os.getenv("OVERRIDE_CLUSTER_CPU_AVG_QUERY",
-                              f'avg_over_time(sum(irate(container_cpu_usage_seconds_total{{}}[5m]))[1h:])')
+                              f'100 * sum(rate(node_cpu_seconds_total{{mode!="idle"}}[1h])) / sum(machine_cpu_cores{{}})')
         return self._get_query_prometheus_value(query=cpu_query)
 
     def get_cluster_avg_memory(self) -> Optional[float]:
         memory_query = os.getenv("OVERRIDE_CLUSTER_MEM_AVG_QUERY",
-                                 f'avg_over_time(sum(container_memory_usage_bytes{{}})[1h:])')
+                                 f'100 * (1 - sum(avg_over_time(node_memory_MemAvailable_bytes{{}}[1h])) / sum(machine_memory_bytes{{}}))')
         return self._get_query_prometheus_value(query=memory_query)
 
     def _get_query_prometheus_value(self, query: str) -> Optional[float]:
-        global_config = self.__global_config
-        prometheus_params = PrometheusParams(**global_config)
-        query_result = run_prometheus_query(prometheus_params=prometheus_params, query=query)
         try:
+            global_config = self.__global_config
+            prometheus_params = PrometheusParams(**global_config)
+            query_result = run_prometheus_query(prometheus_params=prometheus_params, query=query)
             if query_result.result_type == "error" or query_result.vector_result is None:
                 logging.error(f"PrometheusDiscoveryUtils failed to get prometheus results.")
                 return