fix(alloy): use the Kubernetes node name as node_uname_info's nodename

CHANGELOG.md

@@ -10,6 +10,7 @@
 - New `alloy.kubeletVolumeStats` scrapes `kubelet_volume_stats_*` (PVC capacity and usage) from each kubelet, enabled by default. Set `alloy.kubeletVolumeStats.enabled: false` to turn it off.
 - New `alloy.kubeletTLS` configures TLS for the cAdvisor and kubelet scrapes. The defaults (`insecure_skip_verify: true`) match the previous behaviour; set `insecure_skip_verify: false` and `ca_file` to verify the kubelet certificate.
 - Alloy `ClusterRole` is narrowed to `pods`, `nodes` and `nodes/metrics`. It no longer grants read access to `secrets` and `configmaps`. If you add components through `alloy.extraBlocks` that need more access, override `alloy.rbac.rules` and `alloy.rbac.clusterRules`.
+- `node_uname_info` carries the Kubernetes node name in its `nodename` label instead of the OS hostname, so dashboards that match it against `kube_node_info`'s `node` work when nodes are named by instance ID.
 
 ## v0.3.0
 

coder-observability/templates/_collector-config.tpl

@@ -296,6 +296,15 @@ prometheus.relabel "pods" {
     target_label  = "node"
     replacement   = "$1"
   }
+  // node-exporter reports the OS hostname (ip-10-0-0-1.ec2.internal) in node_uname_info's nodename, which does not
+  // match the Kubernetes node name when nodes are named by instance ID. Dashboards join the two to find a node's
+  // instance, so use the node name the pod runs on.
+  rule {
+    source_labels = ["__name__", "node"]
+    regex         = "^node_uname_info;(.+)$"
+    target_label  = "nodename"
+    replacement   = "$1"
+  }
   rule {
     regex         = "^(exported_.*|image_.*|container_id|id|uid)$"
     action        = "labeldrop"