update alloy-mixin to v1.1.1

Signed-off-by: Weifeng Wang <[email protected]>
qclaogui · May 31, 2024 · 685508e · 685508e
1 parent 4d9f507
commit 685508e
Show file tree

Hide file tree

Showing 5 changed files with 27 additions and 25 deletions.
diff --git a/monitoring-mixins/alloy-mixin/deploy/alloy-mixin-alerts.yaml b/monitoring-mixins/alloy-mixin/deploy/alloy-mixin-alerts.yaml
@@ -1,5 +1,5 @@
 groups:
-    - name: clustering
+    - name: alloy_clustering
       rules:
         - alert: ClusterNotConverging
           annotations:
@@ -11,8 +11,8 @@ groups:
             message: Nodes report different number of peers vs. the count of observed Alloy metrics. Some Alloy metrics may be missing or the cluster is in a split brain state.
           expr: |
             sum without (state) (cluster_node_peers) !=
-            on (cluster, namespace) group_left
-            count by (cluster, namespace) (cluster_node_info)
+            on (cluster, namespace, job) group_left
+            count by (cluster, namespace, job) (cluster_node_info)
           for: 15m
         - alert: ClusterNodeUnhealthy
           annotations:
@@ -35,7 +35,7 @@ groups:
             message: Cluster nodes are not using the same configuration file.
           expr: |
             count without (sha256) (
-                max by (cluster, namespace, sha256) (alloy_config_hash and on(cluster, namespace) cluster_node_info)
+                max by (cluster, namespace, sha256, job) (alloy_config_hash and on(cluster, namespace) cluster_node_info)
             ) > 1
           for: 5m
     - name: alloy_controller
@@ -50,15 +50,15 @@ groups:
             message: Unhealthy components detected.
           expr: sum by (cluster, namespace) (alloy_component_controller_running_components{health_type!="healthy"}) > 0
           for: 15m
-    - name: otelcol
+    - name: alloy_otelcol
       rules:
         - alert: OtelcolReceiverRefusedSpans
           annotations:
             message: The receiver could not push some spans to the pipeline.
-          expr: sum(rate(receiver_refused_spans_ratio_total{}[1m])) > 0
+          expr: sum by (cluster, namespace) (rate(receiver_refused_spans_ratio_total{}[1m])) > 0
           for: 5m
         - alert: OtelcolExporterFailedSpans
           annotations:
             message: The exporter failed to send spans to their destination.
-          expr: sum(rate(exporter_send_failed_spans_ratio_total{}[1m])) > 0
+          expr: sum by (cluster, namespace) (rate(exporter_send_failed_spans_ratio_total{}[1m])) > 0
           for: 5m
diff --git a/monitoring-mixins/alloy-mixin/deploy/manifests/k8s-all-in-one.yaml b/monitoring-mixins/alloy-mixin/deploy/manifests/k8s-all-in-one.yaml
@@ -2847,11 +2847,11 @@ metadata:
 apiVersion: monitoring.coreos.com/v1
 kind: PrometheusRule
 metadata:
-  name: agent-flow-mixin-alerts
+  name: alloy-mixin-alerts
   namespace: monitoring-system
 spec:
   groups:
-  - name: clustering
+  - name: alloy_clustering
     rules:
     - alert: ClusterNotConverging
       annotations:
@@ -2867,8 +2867,8 @@ spec:
           split brain state.
       expr: |
         sum without (state) (cluster_node_peers) !=
-        on (cluster, namespace) group_left
-        count by (cluster, namespace) (cluster_node_info)
+        on (cluster, namespace, job) group_left
+        count by (cluster, namespace, job) (cluster_node_info)
       for: 15m
     - alert: ClusterNodeUnhealthy
       annotations:
@@ -2894,7 +2894,7 @@ spec:
         message: Cluster nodes are not using the same configuration file.
       expr: |
         count without (sha256) (
-            max by (cluster, namespace, sha256) (alloy_config_hash and on(cluster, namespace) cluster_node_info)
+            max by (cluster, namespace, sha256, job) (alloy_config_hash and on(cluster, namespace) cluster_node_info)
         ) > 1
       for: 5m
   - name: alloy_controller
@@ -2911,15 +2911,17 @@ spec:
       expr: sum by (cluster, namespace) (alloy_component_controller_running_components{health_type!="healthy"})
         > 0
       for: 15m
-  - name: otelcol
+  - name: alloy_otelcol
     rules:
     - alert: OtelcolReceiverRefusedSpans
       annotations:
         message: The receiver could not push some spans to the pipeline.
-      expr: sum(rate(receiver_refused_spans_ratio_total{}[1m])) > 0
+      expr: sum by (cluster, namespace) (rate(receiver_refused_spans_ratio_total{}[1m]))
+        > 0
       for: 5m
     - alert: OtelcolExporterFailedSpans
       annotations:
         message: The exporter failed to send spans to their destination.
-      expr: sum(rate(exporter_send_failed_spans_ratio_total{}[1m])) > 0
+      expr: sum by (cluster, namespace) (rate(exporter_send_failed_spans_ratio_total{}[1m]))
+        > 0
       for: 5m
diff --git a/monitoring-mixins/alloy-mixin/deploy/prometheus-alerts.yaml b/monitoring-mixins/alloy-mixin/deploy/prometheus-alerts.yaml
@@ -6,7 +6,7 @@ metadata:
   namespace: monitoring-system
 spec:
   groups:
-    - name: clustering
+    - name: alloy_clustering
       rules:
         - alert: ClusterNotConverging
           annotations:
@@ -18,8 +18,8 @@ spec:
             message: Nodes report different number of peers vs. the count of observed Alloy metrics. Some Alloy metrics may be missing or the cluster is in a split brain state.
           expr: |
             sum without (state) (cluster_node_peers) !=
-            on (cluster, namespace) group_left
-            count by (cluster, namespace) (cluster_node_info)
+            on (cluster, namespace, job) group_left
+            count by (cluster, namespace, job) (cluster_node_info)
           for: 15m
         - alert: ClusterNodeUnhealthy
           annotations:
@@ -42,7 +42,7 @@ spec:
             message: Cluster nodes are not using the same configuration file.
           expr: |
             count without (sha256) (
-                max by (cluster, namespace, sha256) (alloy_config_hash and on(cluster, namespace) cluster_node_info)
+                max by (cluster, namespace, sha256, job) (alloy_config_hash and on(cluster, namespace) cluster_node_info)
             ) > 1
           for: 5m
     - name: alloy_controller
@@ -57,15 +57,15 @@ spec:
             message: Unhealthy components detected.
           expr: sum by (cluster, namespace) (alloy_component_controller_running_components{health_type!="healthy"}) > 0
           for: 15m
-    - name: otelcol
+    - name: alloy_otelcol
       rules:
         - alert: OtelcolReceiverRefusedSpans
           annotations:
             message: The receiver could not push some spans to the pipeline.
-          expr: sum(rate(receiver_refused_spans_ratio_total{}[1m])) > 0
+          expr: sum by (cluster, namespace) (rate(receiver_refused_spans_ratio_total{}[1m])) > 0
           for: 5m
         - alert: OtelcolExporterFailedSpans
           annotations:
             message: The exporter failed to send spans to their destination.
-          expr: sum(rate(exporter_send_failed_spans_ratio_total{}[1m])) > 0
+          expr: sum by (cluster, namespace) (rate(exporter_send_failed_spans_ratio_total{}[1m])) > 0
           for: 5m
diff --git a/monitoring-mixins/alloy-mixin/jsonnetfile.json b/monitoring-mixins/alloy-mixin/jsonnetfile.json
@@ -8,7 +8,7 @@
           "subdir": "operations/alloy-mixin"
         }
       },
-      "version": "v1.0.0"
+      "version": "v1.1.1"
     }
   ],
   "legacyImports": true

diff --git a/monitoring-mixins/alloy-mixin/jsonnetfile.lock.json b/monitoring-mixins/alloy-mixin/jsonnetfile.lock.json
@@ -8,8 +8,8 @@
           "subdir": "operations/alloy-mixin"
         }
       },
-      "version": "acd9c2a8735f32b731c4aae61f62d93ec51c112e",
-      "sum": "8P4ZtEkF86ibmgFxWEuA5T7yM46pcYpG3opzale1W0U="
+      "version": "2307e103dafa62bb58b70bd3efd41b91daaed22e",
+      "sum": "S6iPjJ1bt6s2YtjuOJDoEDlpO2IcnVs8vvkff/oEt1c="
     }
   ],
   "legacyImports": false