Update Telemeter metrics allow-list

Signed-off-by: Simon Pasquier <[email protected]>
rhobs · Oct 16, 2024 · e942112 · e942112
1 parent 62555bd
commit e942112
Show file tree

Hide file tree

Showing 3 changed files with 42 additions and 7 deletions.
diff --git a/Makefile b/Makefile
@@ -107,7 +107,9 @@ resources/observability/grafana/observatorium-logs: format observability/grafana
 	$(JSONNET) -J "$(JSONNET_VENDOR_DIR)" -m resources/observability/grafana/observatorium-logs observability/grafana-obs-logs.jsonnet | $(XARGS) -I{} sh -c 'cat {} | $(GOJSONTOYAML) > {}.yaml' -- {}
 
 .PHONY: whitelisted_metrics
-whitelisted_metrics: $(GOJSONTOYAML) $(GOJQ)
+whitelisted_metrics: $(GOJSONTOYAML) $(GOJQ) configuration/telemeter/metrics.json resources/services/telemeter-template.yaml
+
+configuration/telemeter/metrics.json:
 	@echo ">>>>> Running whitelisted_metrics"
 	# Download the latest metrics file to extract the new added metrics.
 	# NOTE: Because old clusters could still send metrics the whitelisting is append only

diff --git a/configuration/telemeter/metrics.json b/configuration/telemeter/metrics.json
@@ -1,5 +1,6 @@
 [
   "{__name__=\":apiserver_v1_image_imports:sum\"}",
+  "{__name__=\"ALERTS\",alertstate=\"firing\",severity=~\"critical|warning|info|none\"}",
   "{__name__=\"ALERTS\",alertstate=\"firing\"}",
   "{__name__=\"acm_console_page_count:sum\", page=~\"overview-classic|overview-fleet|search|search-details|clusters|application|governance\"}",
   "{__name__=\"acm_managed_cluster_info\"}",
@@ -73,6 +74,8 @@
   "{__name__=\"cluster_installer\"}",
   "{__name__=\"cluster_legacy_scheduler_policy\"}",
   "{__name__=\"cluster_master_schedulable\"}",
+  "{__name__=\"cluster_name:hypershift_nodepools_available_replicas:sum\"}",
+  "{__name__=\"cluster_name:hypershift_nodepools_size:sum\"}",
   "{__name__=\"cluster_operator_conditions\"}",
   "{__name__=\"cluster_operator_up\"}",
   "{__name__=\"cluster_version\"}",
@@ -88,6 +91,7 @@
   "{__name__=\"count:up1\"}",
   "{__name__=\"csv_abnormal\"}",
   "{__name__=\"csv_succeeded\"}",
+  "{__name__=\"enabled:tempo_operator_tempostack_jaeger_ui:sum\",enabled=\"true|false\"}",
   "{__name__=\"eo_es_cluster_management_state_info\"}",
   "{__name__=\"eo_es_defined_delete_namespaces_total\"}",
   "{__name__=\"eo_es_misconfigured_memory_resources_info\"}",
@@ -164,8 +168,6 @@
   "{__name__=\"os_image_url_override:sum\"}",
   "{__name__=\"platform:hypershift_hostedclusters:max\"}",
   "{__name__=\"platform:hypershift_nodepools:max\"}",
-  "{__name__=\"cluster_name:hypershift_nodepools_size:sum\"}",
-  "{__name__=\"cluster_name:hypershift_nodepools_available_replicas:sum\"}",
   "{__name__=\"pod:eo_es_shards_total:max\"}",
   "{__name__=\"profile:cluster_monitoring_operator_collection_profile:max\"}",
   "{__name__=\"rhacs:telemetry:rox_central_info\"}",
@@ -180,8 +182,17 @@
   "{__name__=\"rhods_total_users\"}",
   "{__name__=\"state:rhoam_critical_alerts:max\"}",
   "{__name__=\"state:rhoam_warning_alerts:max\"}",
+  "{__name__=\"state:tempo_operator_tempostack_managed:sum\",state=~\"Managed|Unmanaged\"}",
   "{__name__=\"status:upgrading:version:rhoam_state:max\"}",
   "{__name__=\"subscription_sync_total\"}",
+  "{__name__=\"type:opentelemetry_collector_connectors:sum\",type=\"spanmetricsconnector|forwardconnector\"}",
+  "{__name__=\"type:opentelemetry_collector_exporters:sum\",type=\"debugexporter|loggingexporter|otlpexporter|otlphttpexporter|prometheusexporter|lokiexporter|kafkaexporter|awscloudwatchlogsexporter|loadbalancingexporter\"}",
+  "{__name__=\"type:opentelemetry_collector_extensions:sum\",type=\"zpagesextension|ballastextension|memorylimiterextension|jaegerremotesampling|healthcheckextension|pprofextension|oauth2clientauthextension|oidcauthextension|bearertokenauthextension|filestorage\"}",
+  "{__name__=\"type:opentelemetry_collector_info:sum\",type=\"deployment|daemonset|sidecar|statefulset\"}",
+  "{__name__=\"type:opentelemetry_collector_processors:sum\",type=\"batchprocessor|memorylimiterprocessor|attributesprocessor|resourceprocessor|spanprocessor|k8sattributesprocessor|resourcedetectionprocessor|filterprocessor|routingprocessor|cumulativetodeltaprocessor|groupbyattrsprocessor\"}",
+  "{__name__=\"type:opentelemetry_collector_receivers:sum\",type=\"jaegerreceiver|hostmetricsreceiver|opencensusreceiver|prometheusreceiver|zipkinreceiver|kafkareceiver|filelogreceiver|journaldreceiver|k8seventsreceiver|kubeletstatsreceiver|k8sclusterreceiver|k8sobjectsreceiver\"}",
+  "{__name__=\"type:tempo_operator_tempostack_multi_tenancy:sum\",type=~\"enabled|disabled\"}",
+  "{__name__=\"type:tempo_operator_tempostack_storage_backend:sum\",type=~\"azure|gcs|s3\"}",
   "{__name__=\"up\"}",
   "{__name__=\"visual_web_terminal_sessions_total\"}",
   "{__name__=\"workload:cpu_usage_cores:sum\"}",

diff --git a/resources/services/telemeter-template.yaml b/resources/services/telemeter-template.yaml
@@ -94,6 +94,7 @@ objects:
           - --memcached=memcached-1.memcached.${NAMESPACE}.svc.cluster.local:11211
           - --memcached=memcached-2.memcached.${NAMESPACE}.svc.cluster.local:11211
           - --whitelist={__name__=":apiserver_v1_image_imports:sum"}
+          - --whitelist={__name__="alerts",alertstate="firing",severity=~"critical|warning|info|none"}
           - --whitelist={__name__="alerts",alertstate="firing"}
           - --whitelist={__name__="acm_console_page_count:sum", page=~"overview-classic|overview-fleet|search|search-details|clusters|application|governance"}
           - --whitelist={__name__="acm_managed_cluster_info"}
@@ -167,6 +168,8 @@ objects:
           - --whitelist={__name__="cluster_installer"}
           - --whitelist={__name__="cluster_legacy_scheduler_policy"}
           - --whitelist={__name__="cluster_master_schedulable"}
+          - --whitelist={__name__="cluster_name:hypershift_nodepools_available_replicas:sum"}
+          - --whitelist={__name__="cluster_name:hypershift_nodepools_size:sum"}
           - --whitelist={__name__="cluster_operator_conditions"}
           - --whitelist={__name__="cluster_operator_up"}
           - --whitelist={__name__="cluster_version"}
@@ -182,6 +185,7 @@ objects:
           - --whitelist={__name__="count:up1"}
           - --whitelist={__name__="csv_abnormal"}
           - --whitelist={__name__="csv_succeeded"}
+          - --whitelist={__name__="enabled:tempo_operator_tempostack_jaeger_ui:sum",enabled="true|false"}
           - --whitelist={__name__="eo_es_cluster_management_state_info"}
           - --whitelist={__name__="eo_es_defined_delete_namespaces_total"}
           - --whitelist={__name__="eo_es_misconfigured_memory_resources_info"}
@@ -258,8 +262,6 @@ objects:
           - --whitelist={__name__="os_image_url_override:sum"}
           - --whitelist={__name__="platform:hypershift_hostedclusters:max"}
           - --whitelist={__name__="platform:hypershift_nodepools:max"}
-          - --whitelist={__name__="cluster_name:hypershift_nodepools_size:sum"}
-          - --whitelist={__name__="cluster_name:hypershift_nodepools_available_replicas:sum"}
           - --whitelist={__name__="pod:eo_es_shards_total:max"}
           - --whitelist={__name__="profile:cluster_monitoring_operator_collection_profile:max"}
           - --whitelist={__name__="rhacs:telemetry:rox_central_info"}
@@ -274,8 +276,17 @@ objects:
           - --whitelist={__name__="rhods_total_users"}
           - --whitelist={__name__="state:rhoam_critical_alerts:max"}
           - --whitelist={__name__="state:rhoam_warning_alerts:max"}
+          - --whitelist={__name__="state:tempo_operator_tempostack_managed:sum",state=~"Managed|Unmanaged"}
           - --whitelist={__name__="status:upgrading:version:rhoam_state:max"}
           - --whitelist={__name__="subscription_sync_total"}
+          - --whitelist={__name__="type:opentelemetry_collector_connectors:sum",type="spanmetricsconnector|forwardconnector"}
+          - --whitelist={__name__="type:opentelemetry_collector_exporters:sum",type="debugexporter|loggingexporter|otlpexporter|otlphttpexporter|prometheusexporter|lokiexporter|kafkaexporter|awscloudwatchlogsexporter|loadbalancingexporter"}
+          - --whitelist={__name__="type:opentelemetry_collector_extensions:sum",type="zpagesextension|ballastextension|memorylimiterextension|jaegerremotesampling|healthcheckextension|pprofextension|oauth2clientauthextension|oidcauthextension|bearertokenauthextension|filestorage"}
+          - --whitelist={__name__="type:opentelemetry_collector_info:sum",type="deployment|daemonset|sidecar|statefulset"}
+          - --whitelist={__name__="type:opentelemetry_collector_processors:sum",type="batchprocessor|memorylimiterprocessor|attributesprocessor|resourceprocessor|spanprocessor|k8sattributesprocessor|resourcedetectionprocessor|filterprocessor|routingprocessor|cumulativetodeltaprocessor|groupbyattrsprocessor"}
+          - --whitelist={__name__="type:opentelemetry_collector_receivers:sum",type="jaegerreceiver|hostmetricsreceiver|opencensusreceiver|prometheusreceiver|zipkinreceiver|kafkareceiver|filelogreceiver|journaldreceiver|k8seventsreceiver|kubeletstatsreceiver|k8sclusterreceiver|k8sobjectsreceiver"}
+          - --whitelist={__name__="type:tempo_operator_tempostack_multi_tenancy:sum",type=~"enabled|disabled"}
+          - --whitelist={__name__="type:tempo_operator_tempostack_storage_backend:sum",type=~"azure|gcs|s3"}
           - --whitelist={__name__="up"}
           - --whitelist={__name__="visual_web_terminal_sessions_total"}
           - --whitelist={__name__="workload:cpu_usage_cores:sum"}
@@ -379,6 +390,7 @@ objects:
           - --memcached=memcached-1.memcached.${NAMESPACE}.svc.cluster.local:11211
           - --memcached=memcached-2.memcached.${NAMESPACE}.svc.cluster.local:11211
           - --whitelist={__name__=":apiserver_v1_image_imports:sum"}
+          - --whitelist={__name__="alerts",alertstate="firing",severity=~"critical|warning|info|none"}
           - --whitelist={__name__="alerts",alertstate="firing"}
           - --whitelist={__name__="acm_console_page_count:sum", page=~"overview-classic|overview-fleet|search|search-details|clusters|application|governance"}
           - --whitelist={__name__="acm_managed_cluster_info"}
@@ -452,6 +464,8 @@ objects:
           - --whitelist={__name__="cluster_installer"}
           - --whitelist={__name__="cluster_legacy_scheduler_policy"}
           - --whitelist={__name__="cluster_master_schedulable"}
+          - --whitelist={__name__="cluster_name:hypershift_nodepools_available_replicas:sum"}
+          - --whitelist={__name__="cluster_name:hypershift_nodepools_size:sum"}
           - --whitelist={__name__="cluster_operator_conditions"}
           - --whitelist={__name__="cluster_operator_up"}
           - --whitelist={__name__="cluster_version"}
@@ -467,6 +481,7 @@ objects:
           - --whitelist={__name__="count:up1"}
           - --whitelist={__name__="csv_abnormal"}
           - --whitelist={__name__="csv_succeeded"}
+          - --whitelist={__name__="enabled:tempo_operator_tempostack_jaeger_ui:sum",enabled="true|false"}
           - --whitelist={__name__="eo_es_cluster_management_state_info"}
           - --whitelist={__name__="eo_es_defined_delete_namespaces_total"}
           - --whitelist={__name__="eo_es_misconfigured_memory_resources_info"}
@@ -543,8 +558,6 @@ objects:
           - --whitelist={__name__="os_image_url_override:sum"}
           - --whitelist={__name__="platform:hypershift_hostedclusters:max"}
           - --whitelist={__name__="platform:hypershift_nodepools:max"}
-          - --whitelist={__name__="cluster_name:hypershift_nodepools_size:sum"}
-          - --whitelist={__name__="cluster_name:hypershift_nodepools_available_replicas:sum"}
           - --whitelist={__name__="pod:eo_es_shards_total:max"}
           - --whitelist={__name__="profile:cluster_monitoring_operator_collection_profile:max"}
           - --whitelist={__name__="rhacs:telemetry:rox_central_info"}
@@ -559,8 +572,17 @@ objects:
           - --whitelist={__name__="rhods_total_users"}
           - --whitelist={__name__="state:rhoam_critical_alerts:max"}
           - --whitelist={__name__="state:rhoam_warning_alerts:max"}
+          - --whitelist={__name__="state:tempo_operator_tempostack_managed:sum",state=~"Managed|Unmanaged"}
           - --whitelist={__name__="status:upgrading:version:rhoam_state:max"}
           - --whitelist={__name__="subscription_sync_total"}
+          - --whitelist={__name__="type:opentelemetry_collector_connectors:sum",type="spanmetricsconnector|forwardconnector"}
+          - --whitelist={__name__="type:opentelemetry_collector_exporters:sum",type="debugexporter|loggingexporter|otlpexporter|otlphttpexporter|prometheusexporter|lokiexporter|kafkaexporter|awscloudwatchlogsexporter|loadbalancingexporter"}
+          - --whitelist={__name__="type:opentelemetry_collector_extensions:sum",type="zpagesextension|ballastextension|memorylimiterextension|jaegerremotesampling|healthcheckextension|pprofextension|oauth2clientauthextension|oidcauthextension|bearertokenauthextension|filestorage"}
+          - --whitelist={__name__="type:opentelemetry_collector_info:sum",type="deployment|daemonset|sidecar|statefulset"}
+          - --whitelist={__name__="type:opentelemetry_collector_processors:sum",type="batchprocessor|memorylimiterprocessor|attributesprocessor|resourceprocessor|spanprocessor|k8sattributesprocessor|resourcedetectionprocessor|filterprocessor|routingprocessor|cumulativetodeltaprocessor|groupbyattrsprocessor"}
+          - --whitelist={__name__="type:opentelemetry_collector_receivers:sum",type="jaegerreceiver|hostmetricsreceiver|opencensusreceiver|prometheusreceiver|zipkinreceiver|kafkareceiver|filelogreceiver|journaldreceiver|k8seventsreceiver|kubeletstatsreceiver|k8sclusterreceiver|k8sobjectsreceiver"}
+          - --whitelist={__name__="type:tempo_operator_tempostack_multi_tenancy:sum",type=~"enabled|disabled"}
+          - --whitelist={__name__="type:tempo_operator_tempostack_storage_backend:sum",type=~"azure|gcs|s3"}
           - --whitelist={__name__="up"}
           - --whitelist={__name__="visual_web_terminal_sessions_total"}
           - --whitelist={__name__="workload:cpu_usage_cores:sum"}