Updating telemeter rules from openshift/telemeter#506 (#682)

Signed-off-by: Moad Zardab <[email protected]>
rhobs · Feb 19, 2024 · 8c97fcd · 8c97fcd
1 parent c469b01
commit 8c97fcd
Show file tree

Hide file tree

Showing 3 changed files with 16 additions and 24 deletions.
diff --git a/jsonnetfile.lock.json b/jsonnetfile.lock.json
@@ -293,8 +293,8 @@
           "subdir": "jsonnet/telemeter"
         }
       },
-      "version": "23b2e4b9213b97f4bc4e82acb7dcb1aab9604c35",
-      "sum": "1p/t4SQjNEyKU+D3Bw17IK/rDzcFSWc0Vk5IPl3TOCo="
+      "version": "a9417a6062c3a31ed78c06ea3a0613a52f2029b2",
+      "sum": "IFJkQRq4+QFi0/7Rc5OVdqyV6KXNXeBp8hyFAcdEna8="
     },
     {
       "source": {

diff --git a/...ices/app-sre-stage-01/rhobs/telemeter/observatorium-metrics-ruler-telemeter-template.yaml b/...ices/app-sre-stage-01/rhobs/telemeter/observatorium-metrics-ruler-telemeter-template.yaml
@@ -125,23 +125,19 @@ objects:
                   tenant_id: FB870BF3-9F3A-44FF-9BF7-D7A047A52F43
                 record: cluster:usage:workload:capacity_virtual_cpu_hours
               - expr: |
-                  group by(_id) (node_role_os_version_machine:cpu_capacity_cores:sum{label_node_openshift_io_os_id="rhcos"}) * 0
+                  group by(_id, tenant_id) (cluster:capacity_cpu_cores:sum{label_node_openshift_io_os_id="rhcos"}) * 0
                 labels:
                   tenant_id: FB870BF3-9F3A-44FF-9BF7-D7A047A52F43
                 record: cluster:cpu_capacity_cores:_id
               - expr: |
-                  # worker ht amd64
-                  (sum by (_id) (node_role_os_version_machine:cpu_capacity_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io_infra!="true",label_node_role_kubernetes_io_master!="true",label_kubernetes_io_arch="amd64",label_node_hyperthread_enabled="true"}) or cluster:cpu_capacity_cores:_id)+
-                  # worker non-ht amd64
-                  (sum by (_id) (node_role_os_version_machine:cpu_capacity_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io_infra!="true",label_node_role_kubernetes_io_master!="true",label_kubernetes_io_arch="amd64",label_node_hyperthread_enabled="false"}) / 2.0 or cluster:cpu_capacity_cores:_id)+
+                  # worker amd64
+                  (sum by (_id, tenant_id) (cluster:capacity_cpu_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io!="master",label_node_role_kubernetes_io!="infra",label_kubernetes_io_arch="amd64"}) / 2.0 or cluster:cpu_capacity_cores:_id) +
                   # worker non-amd64
-                  (sum by (_id) (node_role_os_version_machine:cpu_capacity_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io_infra!="true",label_node_role_kubernetes_io_master!="true",label_kubernetes_io_arch!="amd64"}) or cluster:cpu_capacity_cores:_id)+
-                  # schedulable control plane ht amd64
-                  (sum by (_id) (node_role_os_version_machine:cpu_capacity_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io_master="true",label_kubernetes_io_arch="amd64",label_node_hyperthread_enabled="true"}) * on(_id) group by(_id) (cluster_master_schedulable == 1) or cluster:cpu_capacity_cores:_id)+
-                  # schedulable control plane non-ht amd64
-                  (sum by (_id) (node_role_os_version_machine:cpu_capacity_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io_master="true",label_kubernetes_io_arch="amd64",label_node_hyperthread_enabled="false"}) * on(_id) group by(_id) (cluster_master_schedulable == 1) / 2.0 or cluster:cpu_capacity_cores:_id)+
+                  (sum by (_id, tenant_id) (cluster:capacity_cpu_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io!="master",label_node_role_kubernetes_io!="infra",label_kubernetes_io_arch!="amd64"}) or cluster:cpu_capacity_cores:_id) +
+                  # schedulable control plane amd64
+                  (sum by (_id, tenant_id) (cluster:capacity_cpu_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io="master",label_kubernetes_io_arch="amd64"}) * on(_id, tenant_id) group by(_id, tenant_id) (cluster_master_schedulable == 1) / 2.0 or cluster:cpu_capacity_cores:_id) +
                   # schedulable control plane non-amd64
-                  (sum by (_id) (node_role_os_version_machine:cpu_capacity_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io_master="true",label_kubernetes_io_arch!="amd64"}) * on(_id) group by(_id) (cluster_master_schedulable == 1) or cluster:cpu_capacity_cores:_id)
+                  (sum by (_id, tenant_id) (cluster:capacity_cpu_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io="master",label_kubernetes_io_arch!="amd64"}) * on(_id, tenant_id) group by(_id, tenant_id) (cluster_master_schedulable == 1) or cluster:cpu_capacity_cores:_id)
                 labels:
                   tenant_id: FB870BF3-9F3A-44FF-9BF7-D7A047A52F43
                 record: cluster:capacity_effective_cpu_cores

diff --git a/resources/services/observatorium-metrics-template.yaml b/resources/services/observatorium-metrics-template.yaml
@@ -2140,23 +2140,19 @@ objects:
             "tenant_id": "FB870BF3-9F3A-44FF-9BF7-D7A047A52F43"
           "record": "cluster:usage:workload:capacity_virtual_cpu_hours"
         - "expr": |
-            group by(_id) (node_role_os_version_machine:cpu_capacity_cores:sum{label_node_openshift_io_os_id="rhcos"}) * 0
+            group by(_id, tenant_id) (cluster:capacity_cpu_cores:sum{label_node_openshift_io_os_id="rhcos"}) * 0
           "labels":
             "tenant_id": "FB870BF3-9F3A-44FF-9BF7-D7A047A52F43"
           "record": "cluster:cpu_capacity_cores:_id"
         - "expr": |
-            # worker ht amd64
-            (sum by (_id) (node_role_os_version_machine:cpu_capacity_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io_infra!="true",label_node_role_kubernetes_io_master!="true",label_kubernetes_io_arch="amd64",label_node_hyperthread_enabled="true"}) or cluster:cpu_capacity_cores:_id)+
-            # worker non-ht amd64
-            (sum by (_id) (node_role_os_version_machine:cpu_capacity_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io_infra!="true",label_node_role_kubernetes_io_master!="true",label_kubernetes_io_arch="amd64",label_node_hyperthread_enabled="false"}) / 2.0 or cluster:cpu_capacity_cores:_id)+
+            # worker amd64
+            (sum by (_id, tenant_id) (cluster:capacity_cpu_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io!="master",label_node_role_kubernetes_io!="infra",label_kubernetes_io_arch="amd64"}) / 2.0 or cluster:cpu_capacity_cores:_id) +
             # worker non-amd64
-            (sum by (_id) (node_role_os_version_machine:cpu_capacity_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io_infra!="true",label_node_role_kubernetes_io_master!="true",label_kubernetes_io_arch!="amd64"}) or cluster:cpu_capacity_cores:_id)+
-            # schedulable control plane ht amd64
-            (sum by (_id) (node_role_os_version_machine:cpu_capacity_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io_master="true",label_kubernetes_io_arch="amd64",label_node_hyperthread_enabled="true"}) * on(_id) group by(_id) (cluster_master_schedulable == 1) or cluster:cpu_capacity_cores:_id)+
-            # schedulable control plane non-ht amd64
-            (sum by (_id) (node_role_os_version_machine:cpu_capacity_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io_master="true",label_kubernetes_io_arch="amd64",label_node_hyperthread_enabled="false"}) * on(_id) group by(_id) (cluster_master_schedulable == 1) / 2.0 or cluster:cpu_capacity_cores:_id)+
+            (sum by (_id, tenant_id) (cluster:capacity_cpu_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io!="master",label_node_role_kubernetes_io!="infra",label_kubernetes_io_arch!="amd64"}) or cluster:cpu_capacity_cores:_id) +
+            # schedulable control plane amd64
+            (sum by (_id, tenant_id) (cluster:capacity_cpu_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io="master",label_kubernetes_io_arch="amd64"}) * on(_id, tenant_id) group by(_id, tenant_id) (cluster_master_schedulable == 1) / 2.0 or cluster:cpu_capacity_cores:_id) +
             # schedulable control plane non-amd64
-            (sum by (_id) (node_role_os_version_machine:cpu_capacity_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io_master="true",label_kubernetes_io_arch!="amd64"}) * on(_id) group by(_id) (cluster_master_schedulable == 1) or cluster:cpu_capacity_cores:_id)
+            (sum by (_id, tenant_id) (cluster:capacity_cpu_cores:sum{label_node_openshift_io_os_id="rhcos",label_node_role_kubernetes_io="master",label_kubernetes_io_arch!="amd64"}) * on(_id, tenant_id) group by(_id, tenant_id) (cluster_master_schedulable == 1) or cluster:cpu_capacity_cores:_id)
           "labels":
             "tenant_id": "FB870BF3-9F3A-44FF-9BF7-D7A047A52F43"
           "record": "cluster:capacity_effective_cpu_cores"