changed promqls and fixed several monitoring bugs
This commit is contained in:
Carman Zhang
@@ -89,9 +89,9 @@ var ClusterMetricsNames = []string{
|
||||
"cluster_cpu_usage",
|
||||
"cluster_cpu_total",
|
||||
"cluster_memory_utilisation",
|
||||
"cluster_memory_bytes_available",
|
||||
"cluster_memory_bytes_total",
|
||||
"cluster_memory_bytes_usage",
|
||||
"cluster_memory_available",
|
||||
"cluster_memory_total",
|
||||
"cluster_memory_usage_wo_cache",
|
||||
"cluster_net_utilisation",
|
||||
"cluster_net_bytes_transmitted",
|
||||
"cluster_net_bytes_received",
|
||||
@@ -131,7 +131,6 @@ var ClusterMetricsNames = []string{
|
||||
"cluster_secret_count",
|
||||
|
||||
"cluster_namespace_count",
|
||||
"workspace_all_project_count",
|
||||
|
||||
"cluster_load1",
|
||||
"cluster_load5",
|
||||
@@ -142,9 +141,10 @@ var NodeMetricsNames = []string{
|
||||
"node_cpu_total",
|
||||
"node_cpu_usage",
|
||||
"node_memory_utilisation",
|
||||
"node_memory_bytes_usage",
|
||||
"node_memory_bytes_available",
|
||||
"node_memory_bytes_total",
|
||||
"node_memory_usage_wo_cache",
|
||||
"node_memory_available",
|
||||
"node_memory_total",
|
||||
|
||||
"node_net_utilisation",
|
||||
"node_net_bytes_transmitted",
|
||||
"node_net_bytes_received",
|
||||
@@ -195,6 +195,8 @@ var WorkspaceMetricsNames = []string{
|
||||
"workspace_replicaset_count",
|
||||
"workspace_service_count",
|
||||
"workspace_secret_count",
|
||||
|
||||
"workspace_all_project_count",
|
||||
}
|
||||
var NamespaceMetricsNames = []string{
|
||||
"namespace_cpu_usage",
|
||||
@@ -280,31 +282,31 @@ var WorkloadMetricsNames = []string{
|
||||
|
||||
var RulePromQLTmplMap = MetricMap{
|
||||
//cluster
|
||||
"cluster_cpu_utilisation": ":node_cpu_utilisation:avg1m",
|
||||
"cluster_cpu_usage": `:node_cpu_utilisation:avg1m * sum(node:node_num_cpu:sum)`,
|
||||
"cluster_cpu_total": "sum(node:node_num_cpu:sum)",
|
||||
"cluster_memory_utilisation": ":node_memory_utilisation:",
|
||||
"cluster_memory_bytes_available": "sum(node:node_memory_bytes_available:sum)",
|
||||
"cluster_memory_bytes_total": "sum(node:node_memory_bytes_total:sum)",
|
||||
"cluster_memory_bytes_usage": "sum(node:node_memory_bytes_total:sum) - sum(node:node_memory_bytes_available:sum)",
|
||||
"cluster_net_utilisation": ":node_net_utilisation:sum_irate",
|
||||
"cluster_net_bytes_transmitted": "sum(node:node_net_bytes_transmitted:sum_irate)",
|
||||
"cluster_net_bytes_received": "sum(node:node_net_bytes_received:sum_irate)",
|
||||
"cluster_disk_read_iops": "sum(node:data_volume_iops_reads:sum)",
|
||||
"cluster_disk_write_iops": "sum(node:data_volume_iops_writes:sum)",
|
||||
"cluster_disk_read_throughput": "sum(node:data_volume_throughput_bytes_read:sum)",
|
||||
"cluster_disk_write_throughput": "sum(node:data_volume_throughput_bytes_written:sum)",
|
||||
"cluster_disk_size_usage": `sum(sum by (node) ((node_filesystem_size{mountpoint="/", job="node-exporter"}) * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:)) - sum(sum by (node) ((node_filesystem_avail{mountpoint="/", job="node-exporter"}) * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:))`,
|
||||
"cluster_disk_size_utilisation": `(sum(sum by (node) ((node_filesystem_size{mountpoint="/", job="node-exporter"}) * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:)) - sum(sum by (node) ((node_filesystem_avail{mountpoint="/", job="node-exporter"}) * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:))) / sum(sum by (node) ((node_filesystem_size{mountpoint="/", job="node-exporter"}) * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:))`,
|
||||
"cluster_disk_size_capacity": `sum(sum by (node) ((node_filesystem_size{mountpoint="/", job="node-exporter"}) * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:))`,
|
||||
"cluster_disk_size_available": `sum(sum by (node) ((node_filesystem_avail{mountpoint="/", job="node-exporter"}) * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:))`,
|
||||
"cluster_cpu_utilisation": ":node_cpu_utilisation:avg1m",
|
||||
"cluster_cpu_usage": `:node_cpu_utilisation:avg1m * sum(node:node_num_cpu:sum)`,
|
||||
"cluster_cpu_total": "sum(node:node_num_cpu:sum)",
|
||||
"cluster_memory_utilisation": ":node_memory_utilisation:",
|
||||
"cluster_memory_available": "sum(node:node_memory_bytes_available:sum)",
|
||||
"cluster_memory_total": "sum(node:node_memory_bytes_total:sum)",
|
||||
"cluster_memory_usage_wo_cache": "sum(node:node_memory_bytes_total:sum) - sum(node:node_memory_bytes_available:sum)",
|
||||
|
||||
"cluster_net_utilisation": ":node_net_utilisation:sum_irate",
|
||||
"cluster_net_bytes_transmitted": "sum(node:node_net_bytes_transmitted:sum_irate)",
|
||||
"cluster_net_bytes_received": "sum(node:node_net_bytes_received:sum_irate)",
|
||||
"cluster_disk_read_iops": "sum(node:data_volume_iops_reads:sum)",
|
||||
"cluster_disk_write_iops": "sum(node:data_volume_iops_writes:sum)",
|
||||
"cluster_disk_read_throughput": "sum(node:data_volume_throughput_bytes_read:sum)",
|
||||
"cluster_disk_write_throughput": "sum(node:data_volume_throughput_bytes_written:sum)",
|
||||
"cluster_disk_size_usage": `sum(sum by (node) ((node_filesystem_size{mountpoint="/", job="node-exporter"}) * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:)) - sum(sum by (node) ((node_filesystem_avail{mountpoint="/", job="node-exporter"}) * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:))`,
|
||||
"cluster_disk_size_utilisation": `(sum(sum by (node) ((node_filesystem_size{mountpoint="/", job="node-exporter"}) * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:)) - sum(sum by (node) ((node_filesystem_avail{mountpoint="/", job="node-exporter"}) * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:))) / sum(sum by (node) ((node_filesystem_size{mountpoint="/", job="node-exporter"}) * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:))`,
|
||||
"cluster_disk_size_capacity": `sum(sum by (node) ((node_filesystem_size{mountpoint="/", job="node-exporter"}) * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:))`,
|
||||
"cluster_disk_size_available": `sum(sum by (node) ((node_filesystem_avail{mountpoint="/", job="node-exporter"}) * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:))`,
|
||||
|
||||
"cluster_disk_inode_total": `sum(node:disk_inodes_total:)`,
|
||||
"cluster_disk_inode_usage": `sum(node:disk_inodes_total:) - sum(node:disk_inodes_free:)`,
|
||||
"cluster_disk_inode_utilisation": `1 - sum(node:disk_inodes_free:) / sum(node:disk_inodes_total:)`,
|
||||
|
||||
"cluster_namespace_count": `count(kube_namespace_annotations)`,
|
||||
"workspace_all_project_count": `count(kube_namespace_annotations)`,
|
||||
"cluster_namespace_count": `count(kube_namespace_annotations)`,
|
||||
|
||||
// cluster_pod_count = cluster_pod_running_count + cluster_pod_succeeded_count + cluster_pod_abnormal_count
|
||||
"cluster_pod_count": `sum(kube_pod_info unless on (node) (kube_node_status_condition{condition="Ready",status=~"unknown|false"} > 0))`,
|
||||
@@ -353,16 +355,18 @@ var RulePromQLTmplMap = MetricMap{
|
||||
"cluster_secret_count": `sum(kube_secret_info)`,
|
||||
"cluster_pv_count": `sum(kube_persistentvolume_labels)`,
|
||||
|
||||
"cluster_load1": `sum(node_load1{job="node-exporter"})`,
|
||||
"cluster_load5": `sum(node_load5{job="node-exporter"})`,
|
||||
"cluster_load15": `sum(node_load15{job="node-exporter"})`,
|
||||
"cluster_load1": `sum(node_load1{job="node-exporter"}) / sum(node:node_num_cpu:sum)`,
|
||||
"cluster_load5": `sum(node_load5{job="node-exporter"}) / sum(node:node_num_cpu:sum)`,
|
||||
"cluster_load15": `sum(node_load15{job="node-exporter"}) / sum(node:node_num_cpu:sum)`,
|
||||
|
||||
//node
|
||||
"node_cpu_utilisation": "node:node_cpu_utilisation:avg1m",
|
||||
"node_cpu_total": "node:node_num_cpu:sum",
|
||||
"node_memory_utilisation": "node:node_memory_utilisation:",
|
||||
"node_memory_bytes_available": "node:node_memory_bytes_available:sum",
|
||||
"node_memory_bytes_total": "node:node_memory_bytes_total:sum",
|
||||
"node_cpu_utilisation": "node:node_cpu_utilisation:avg1m",
|
||||
"node_cpu_total": "node:node_num_cpu:sum",
|
||||
"node_memory_utilisation": "node:node_memory_utilisation:",
|
||||
"node_memory_available": "node:node_memory_bytes_available:sum",
|
||||
"node_memory_total": "node:node_memory_bytes_total:sum",
|
||||
"node_memory_usage_wo_cache": "node:node_memory_bytes_total:sum$1 - node:node_memory_bytes_available:sum$1",
|
||||
|
||||
// Node network utilisation (bytes received + bytes transmitted per second)
|
||||
"node_net_utilisation": "node:node_net_utilisation:sum_irate",
|
||||
// Node network bytes transmitted per second
|
||||
@@ -396,77 +400,76 @@ var RulePromQLTmplMap = MetricMap{
|
||||
"node_pod_abnormal_count": `count(kube_pod_info$1 unless on (pod) (kube_pod_status_phase{phase=~"Succeeded|Running"} > 0)) by (node) unless on (node) (kube_node_status_condition{condition="Ready",status=~"unknown|false"} > 0)`,
|
||||
|
||||
// without log node: unless on(node) kube_node_labels{label_role="log"}
|
||||
"node_cpu_usage": `node:node_cpu_utilisation:avg1m$1 * node:node_num_cpu:sum$1`,
|
||||
"node_memory_bytes_usage": "node:node_memory_bytes_total:sum$1 - node:node_memory_bytes_available:sum$1",
|
||||
"node_cpu_usage": `node:node_cpu_utilisation:avg1m$1 * node:node_num_cpu:sum$1`,
|
||||
|
||||
"node_load1": `sum by (node) (node_load1{job="node-exporter"} * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:$1)`,
|
||||
"node_load5": `sum by (node) (node_load5{job="node-exporter"} * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:$1)`,
|
||||
"node_load15": `sum by (node) (node_load15{job="node-exporter"} * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:$1)`,
|
||||
"node_load1": `sum by (node) (node_load1{job="node-exporter"} * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:$1) / node:node_num_cpu:sum`,
|
||||
"node_load5": `sum by (node) (node_load5{job="node-exporter"} * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:$1) / node:node_num_cpu:sum`,
|
||||
"node_load15": `sum by (node) (node_load15{job="node-exporter"} * on (namespace, pod) group_left(node) node_namespace_pod:kube_pod_info:$1) / node:node_num_cpu:sum`,
|
||||
|
||||
//namespace
|
||||
"namespace_cpu_usage": `namespace:container_cpu_usage_seconds_total:sum_rate{namespace=~"$1"}`,
|
||||
"namespace_memory_usage": `namespace:container_memory_usage_bytes:sum{namespace=~"$1"}`,
|
||||
"namespace_memory_usage_wo_cache": `namespace:container_memory_usage_bytes_wo_cache:sum{namespace=~"$1"}`,
|
||||
"namespace_net_bytes_transmitted": `sum by (namespace) (irate(container_network_transmit_bytes_total{namespace=~"$1", pod_name!="", interface="eth0", job="kubelet"}[5m]))`,
|
||||
"namespace_net_bytes_received": `sum by (namespace) (irate(container_network_receive_bytes_total{namespace=~"$1", pod_name!="", interface="eth0", job="kubelet"}[5m]))`,
|
||||
"namespace_pod_count": `sum(kube_pod_status_phase{namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_pod_running_count": `sum(kube_pod_status_phase{phase="Running", namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_pod_succeeded_count": `sum(kube_pod_status_phase{phase="Succeeded", namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_pod_abnormal_count": `sum(kube_pod_status_phase{phase=~"Failed|Pending|Unknown", namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_cpu_usage": `namespace:container_cpu_usage_seconds_total:sum_rate{namespace!="", namespace=~"$1"}`,
|
||||
"namespace_memory_usage": `namespace:container_memory_usage_bytes:sum{namespace!="", namespace=~"$1"}`,
|
||||
"namespace_memory_usage_wo_cache": `namespace:container_memory_usage_bytes_wo_cache:sum{namespace!="", namespace=~"$1"}`,
|
||||
"namespace_net_bytes_transmitted": `sum by (namespace) (irate(container_network_transmit_bytes_total{namespace!="", namespace=~"$1", pod_name!="", interface="eth0", job="kubelet"}[5m]))`,
|
||||
"namespace_net_bytes_received": `sum by (namespace) (irate(container_network_receive_bytes_total{namespace!="", namespace=~"$1", pod_name!="", interface="eth0", job="kubelet"}[5m]))`,
|
||||
"namespace_pod_count": `sum(kube_pod_status_phase{namespace!="", namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_pod_running_count": `sum(kube_pod_status_phase{phase="Running", namespace!="", namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_pod_succeeded_count": `sum(kube_pod_status_phase{phase="Succeeded", namespace!="", namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_pod_abnormal_count": `sum(kube_pod_status_phase{phase=~"Failed|Pending|Unknown", namespace!="", namespace=~"$1"}) by (namespace)`,
|
||||
|
||||
"namespace_configmap_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/configmaps"}) by (namespace, resource, type)`,
|
||||
"namespace_jobs_batch_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/jobs.batch"}) by (namespace, resource, type)`,
|
||||
"namespace_roles_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/roles.rbac.authorization.k8s.io"}) by (namespace, resource, type)`,
|
||||
"namespace_memory_limit_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="limits.memory"}) by (namespace, resource, type)`,
|
||||
"namespace_pvc_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="persistentvolumeclaims"}) by (namespace, resource, type)`,
|
||||
"namespace_memory_request_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="requests.memory"}) by (namespace, resource, type)`,
|
||||
"namespace_pvc_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/persistentvolumeclaims"}) by (namespace, resource, type)`,
|
||||
"namespace_cronjobs_batch_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/cronjobs.batch"}) by (namespace, resource, type)`,
|
||||
"namespace_ingresses_extensions_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/ingresses.extensions"}) by (namespace, resource, type)`,
|
||||
"namespace_cpu_limit_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="limits.cpu"}) by (namespace, resource, type)`,
|
||||
"namespace_storage_request_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="requests.storage"}) by (namespace, resource, type)`,
|
||||
"namespace_deployment_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/deployments.apps"}) by (namespace, resource, type)`,
|
||||
"namespace_pod_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/pods"}) by (namespace, resource, type)`,
|
||||
"namespace_statefulset_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/statefulsets.apps"}) by (namespace, resource, type)`,
|
||||
"namespace_daemonset_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/daemonsets.apps"}) by (namespace, resource, type)`,
|
||||
"namespace_secret_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/secrets"}) by (namespace, resource, type)`,
|
||||
"namespace_service_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/services"}) by (namespace, resource, type)`,
|
||||
"namespace_cpu_request_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="requests.cpu"}) by (namespace, resource, type)`,
|
||||
"namespace_service_loadbalancer_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="services.loadbalancers"}) by (namespace, resource, type)`,
|
||||
"namespace_configmap_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="count/configmaps"}) by (namespace, resource, type)`,
|
||||
"namespace_jobs_batch_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="count/jobs.batch"}) by (namespace, resource, type)`,
|
||||
"namespace_roles_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="count/roles.rbac.authorization.k8s.io"}) by (namespace, resource, type)`,
|
||||
"namespace_memory_limit_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="limits.memory"}) by (namespace, resource, type)`,
|
||||
"namespace_pvc_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="persistentvolumeclaims"}) by (namespace, resource, type)`,
|
||||
"namespace_memory_request_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="requests.memory"}) by (namespace, resource, type)`,
|
||||
"namespace_pvc_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="count/persistentvolumeclaims"}) by (namespace, resource, type)`,
|
||||
"namespace_cronjobs_batch_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="count/cronjobs.batch"}) by (namespace, resource, type)`,
|
||||
"namespace_ingresses_extensions_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="count/ingresses.extensions"}) by (namespace, resource, type)`,
|
||||
"namespace_cpu_limit_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="limits.cpu"}) by (namespace, resource, type)`,
|
||||
"namespace_storage_request_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="requests.storage"}) by (namespace, resource, type)`,
|
||||
"namespace_deployment_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="count/deployments.apps"}) by (namespace, resource, type)`,
|
||||
"namespace_pod_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="count/pods"}) by (namespace, resource, type)`,
|
||||
"namespace_statefulset_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="count/statefulsets.apps"}) by (namespace, resource, type)`,
|
||||
"namespace_daemonset_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="count/daemonsets.apps"}) by (namespace, resource, type)`,
|
||||
"namespace_secret_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="count/secrets"}) by (namespace, resource, type)`,
|
||||
"namespace_service_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="count/services"}) by (namespace, resource, type)`,
|
||||
"namespace_cpu_request_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="requests.cpu"}) by (namespace, resource, type)`,
|
||||
"namespace_service_loadbalancer_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace=~"$1", resource="services.loadbalancers"}) by (namespace, resource, type)`,
|
||||
|
||||
"namespace_configmap_count_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="count/configmaps"}) by (namespace, resource, type)`,
|
||||
"namespace_jobs_batch_count_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="count/jobs.batch"}) by (namespace, resource, type)`,
|
||||
"namespace_roles_count_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="count/roles.rbac.authorization.k8s.io"}) by (namespace, resource, type)`,
|
||||
"namespace_memory_limit_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="limits.memory"}) by (namespace, resource, type)`,
|
||||
"namespace_pvc_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="persistentvolumeclaims"}) by (namespace, resource, type)`,
|
||||
"namespace_memory_request_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="requests.memory"}) by (namespace, resource, type)`,
|
||||
"namespace_pvc_count_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="count/persistentvolumeclaims"}) by (namespace, resource, type)`,
|
||||
"namespace_cronjobs_batch_count_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="count/cronjobs.batch"}) by (namespace, resource, type)`,
|
||||
"namespace_ingresses_extensions_count_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="count/ingresses.extensions"}) by (namespace, resource, type)`,
|
||||
"namespace_cpu_limit_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="limits.cpu"}) by (namespace, resource, type)`,
|
||||
"namespace_storage_request_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="requests.storage"}) by (namespace, resource, type)`,
|
||||
"namespace_deployment_count_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="count/deployments.apps"}) by (namespace, resource, type)`,
|
||||
"namespace_pod_count_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="count/pods"}) by (namespace, resource, type)`,
|
||||
"namespace_statefulset_count_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="count/statefulsets.apps"}) by (namespace, resource, type)`,
|
||||
"namespace_daemonset_count_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="count/daemonsets.apps"}) by (namespace, resource, type)`,
|
||||
"namespace_secret_count_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="count/secrets"}) by (namespace, resource, type)`,
|
||||
"namespace_service_count_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="count/services"}) by (namespace, resource, type)`,
|
||||
"namespace_cpu_request_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="requests.cpu"}) by (namespace, resource, type)`,
|
||||
"namespace_service_loadbalancer_hard": `sum(kube_resourcequota{resourcequota!="quota", resourcequota!="quota", type="hard", namespace=~"$1", resource="services.loadbalancers"}) by (namespace, resource, type)`,
|
||||
"namespace_configmap_count_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="count/configmaps"}) by (namespace, resource, type)`,
|
||||
"namespace_jobs_batch_count_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="count/jobs.batch"}) by (namespace, resource, type)`,
|
||||
"namespace_roles_count_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="count/roles.rbac.authorization.k8s.io"}) by (namespace, resource, type)`,
|
||||
"namespace_memory_limit_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="limits.memory"}) by (namespace, resource, type)`,
|
||||
"namespace_pvc_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="persistentvolumeclaims"}) by (namespace, resource, type)`,
|
||||
"namespace_memory_request_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="requests.memory"}) by (namespace, resource, type)`,
|
||||
"namespace_pvc_count_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="count/persistentvolumeclaims"}) by (namespace, resource, type)`,
|
||||
"namespace_cronjobs_batch_count_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="count/cronjobs.batch"}) by (namespace, resource, type)`,
|
||||
"namespace_ingresses_extensions_count_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="count/ingresses.extensions"}) by (namespace, resource, type)`,
|
||||
"namespace_cpu_limit_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="limits.cpu"}) by (namespace, resource, type)`,
|
||||
"namespace_storage_request_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="requests.storage"}) by (namespace, resource, type)`,
|
||||
"namespace_deployment_count_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="count/deployments.apps"}) by (namespace, resource, type)`,
|
||||
"namespace_pod_count_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="count/pods"}) by (namespace, resource, type)`,
|
||||
"namespace_statefulset_count_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="count/statefulsets.apps"}) by (namespace, resource, type)`,
|
||||
"namespace_daemonset_count_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="count/daemonsets.apps"}) by (namespace, resource, type)`,
|
||||
"namespace_secret_count_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="count/secrets"}) by (namespace, resource, type)`,
|
||||
"namespace_service_count_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="count/services"}) by (namespace, resource, type)`,
|
||||
"namespace_cpu_request_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="requests.cpu"}) by (namespace, resource, type)`,
|
||||
"namespace_service_loadbalancer_hard": `sum(kube_resourcequota{resourcequota!="quota", type="hard", namespace!="", namespace=~"$1", resource="services.loadbalancers"}) by (namespace, resource, type)`,
|
||||
|
||||
"namespace_cronjob_count": `sum(kube_cronjob_labels{namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_pvc_count": `sum(kube_persistentvolumeclaim_info{namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_daemonset_count": `sum(kube_daemonset_labels{namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_deployment_count": `sum(kube_deployment_labels{namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_endpoint_count": `sum(kube_endpoint_labels{namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_hpa_count": `sum(kube_hpa_labels{namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_job_count": `sum(kube_job_labels{namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_statefulset_count": `sum(kube_statefulset_labels{namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_replicaset_count": `count(kube_replicaset_created{namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_service_count": `sum(kube_service_info{namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_secret_count": `sum(kube_secret_info{namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_cronjob_count": `sum(kube_cronjob_labels{namespace!="", namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_pvc_count": `sum(kube_persistentvolumeclaim_info{namespace!="", namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_daemonset_count": `sum(kube_daemonset_labels{namespace!="", namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_deployment_count": `sum(kube_deployment_labels{namespace!="", namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_endpoint_count": `sum(kube_endpoint_labels{namespace!="", namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_hpa_count": `sum(kube_hpa_labels{namespace!="", namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_job_count": `sum(kube_job_labels{namespace!="", namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_statefulset_count": `sum(kube_statefulset_labels{namespace!="", namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_replicaset_count": `count(kube_replicaset_created{namespace!="", namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_service_count": `sum(kube_service_info{namespace!="", namespace=~"$1"}) by (namespace)`,
|
||||
"namespace_secret_count": `sum(kube_secret_info{namespace!="", namespace=~"$1"}) by (namespace)`,
|
||||
|
||||
"namespace_ingresses_extensions_count": `sum(kube_ingress_labels{namespace=~"$1"})`,
|
||||
"namespace_ingresses_extensions_count": `sum(kube_ingress_labels{namespace!="", namespace=~"$1"}) by (namespace)`,
|
||||
|
||||
// pod
|
||||
"pod_cpu_usage": `sum(irate(container_cpu_usage_seconds_total{job="kubelet", namespace="$1", pod_name="$2", image!=""}[5m])) by (namespace, pod_name)`,
|
||||
@@ -496,47 +499,49 @@ var RulePromQLTmplMap = MetricMap{
|
||||
"container_memory_usage_all": `container_memory_usage_bytes{namespace="$1", pod_name="$2", container_name=~"$3", container_name!="POD"}`,
|
||||
|
||||
// workspace
|
||||
"workspace_cpu_usage": `sum(namespace:container_cpu_usage_seconds_total:sum_rate{namespace =~"$1"})`,
|
||||
"workspace_memory_usage": `sum(namespace:container_memory_usage_bytes:sum{namespace =~"$1"})`,
|
||||
"workspace_memory_usage_wo_cache": `sum(namespace:container_memory_usage_bytes_wo_cache:sum{namespace =~"$1"})`,
|
||||
"workspace_net_bytes_transmitted": `sum(sum by (namespace) (irate(container_network_transmit_bytes_total{namespace=~"$1", pod_name!="", interface="eth0", job="kubelet"}[5m])))`,
|
||||
"workspace_net_bytes_received": `sum(sum by (namespace) (irate(container_network_receive_bytes_total{namespace=~"$1", pod_name!="", interface="eth0", job="kubelet"}[5m])))`,
|
||||
"workspace_pod_count": `sum(kube_pod_status_phase{namespace=~"$1"})`,
|
||||
"workspace_pod_running_count": `sum(kube_pod_status_phase{phase="Running", namespace=~"$1"})`,
|
||||
"workspace_pod_succeeded_count": `sum(kube_pod_status_phase{phase="Succeeded", namespace=~"$1"})`,
|
||||
"workspace_pod_abnormal_count": `sum(kube_pod_status_phase{phase=~"Failed|Pending|Unknown", namespace=~"$1"})`,
|
||||
"workspace_cpu_usage": `sum(namespace:container_cpu_usage_seconds_total:sum_rate{namespace!="", namespace$1})`,
|
||||
"workspace_memory_usage": `sum(namespace:container_memory_usage_bytes:sum{namespace!="", namespace$1})`,
|
||||
"workspace_memory_usage_wo_cache": `sum(namespace:container_memory_usage_bytes_wo_cache:sum{namespace!="", namespace$1})`,
|
||||
"workspace_net_bytes_transmitted": `sum(sum by (namespace) (irate(container_network_transmit_bytes_total{namespace!="", namespace$1, pod_name!="", interface="eth0", job="kubelet"}[5m])))`,
|
||||
"workspace_net_bytes_received": `sum(sum by (namespace) (irate(container_network_receive_bytes_total{namespace!="", namespace$1, pod_name!="", interface="eth0", job="kubelet"}[5m])))`,
|
||||
"workspace_pod_count": `sum(kube_pod_status_phase{namespace!="", namespace$1})`,
|
||||
"workspace_pod_running_count": `sum(kube_pod_status_phase{phase="Running", namespace!="", namespace$1})`,
|
||||
"workspace_pod_succeeded_count": `sum(kube_pod_status_phase{phase="Succeeded", namespace!="", namespace$1})`,
|
||||
"workspace_pod_abnormal_count": `sum(kube_pod_status_phase{phase=~"Failed|Pending|Unknown", namespace!="", namespace$1})`,
|
||||
|
||||
"workspace_configmap_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/configmaps"}) by (resource, type)`,
|
||||
"workspace_jobs_batch_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/jobs.batch"}) by (resource, type)`,
|
||||
"workspace_roles_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/roles.rbac.authorization.k8s.io"}) by (resource, type)`,
|
||||
"workspace_memory_limit_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="limits.memory"}) by (resource, type)`,
|
||||
"workspace_pvc_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="persistentvolumeclaims"}) by (resource, type)`,
|
||||
"workspace_memory_request_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="requests.memory"}) by (resource, type)`,
|
||||
"workspace_pvc_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/persistentvolumeclaims"}) by (resource, type)`,
|
||||
"workspace_cronjobs_batch_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/cronjobs.batch"}) by (resource, type)`,
|
||||
"workspace_ingresses_extensions_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/ingresses.extensions"}) by (resource, type)`,
|
||||
"workspace_cpu_limit_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="limits.cpu"}) by (resource, type)`,
|
||||
"workspace_storage_request_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="requests.storage"}) by (resource, type)`,
|
||||
"workspace_deployment_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/deployments.apps"}) by (resource, type)`,
|
||||
"workspace_pod_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/pods"}) by (resource, type)`,
|
||||
"workspace_statefulset_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/statefulsets.apps"}) by (resource, type)`,
|
||||
"workspace_daemonset_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/daemonsets.apps"}) by (resource, type)`,
|
||||
"workspace_secret_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/secrets"}) by (resource, type)`,
|
||||
"workspace_service_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="count/services"}) by (resource, type)`,
|
||||
"workspace_cpu_request_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="requests.cpu"}) by (resource, type)`,
|
||||
"workspace_service_loadbalancer_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace=~"$1", resource="services.loadbalancers"}) by (resource, type)`,
|
||||
"workspace_configmap_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="count/configmaps"}) by (resource, type)`,
|
||||
"workspace_jobs_batch_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="count/jobs.batch"}) by (resource, type)`,
|
||||
"workspace_roles_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="count/roles.rbac.authorization.k8s.io"}) by (resource, type)`,
|
||||
"workspace_memory_limit_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="limits.memory"}) by (resource, type)`,
|
||||
"workspace_pvc_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="persistentvolumeclaims"}) by (resource, type)`,
|
||||
"workspace_memory_request_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="requests.memory"}) by (resource, type)`,
|
||||
"workspace_pvc_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="count/persistentvolumeclaims"}) by (resource, type)`,
|
||||
"workspace_cronjobs_batch_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="count/cronjobs.batch"}) by (resource, type)`,
|
||||
"workspace_ingresses_extensions_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="count/ingresses.extensions"}) by (resource, type)`,
|
||||
"workspace_cpu_limit_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="limits.cpu"}) by (resource, type)`,
|
||||
"workspace_storage_request_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="requests.storage"}) by (resource, type)`,
|
||||
"workspace_deployment_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="count/deployments.apps"}) by (resource, type)`,
|
||||
"workspace_pod_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="count/pods"}) by (resource, type)`,
|
||||
"workspace_statefulset_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="count/statefulsets.apps"}) by (resource, type)`,
|
||||
"workspace_daemonset_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="count/daemonsets.apps"}) by (resource, type)`,
|
||||
"workspace_secret_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="count/secrets"}) by (resource, type)`,
|
||||
"workspace_service_count_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="count/services"}) by (resource, type)`,
|
||||
"workspace_cpu_request_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="requests.cpu"}) by (resource, type)`,
|
||||
"workspace_service_loadbalancer_used": `sum(kube_resourcequota{resourcequota!="quota", type="used", namespace!="", namespace$1, resource="services.loadbalancers"}) by (resource, type)`,
|
||||
|
||||
"workspace_ingresses_extensions_count": `sum(kube_ingress_labels{namespace=~"$1"})`,
|
||||
"workspace_ingresses_extensions_count": `sum(kube_ingress_labels{namespace!="", namespace$1})`,
|
||||
|
||||
"workspace_cronjob_count": `sum(kube_cronjob_labels{namespace=~"$1"})`,
|
||||
"workspace_pvc_count": `sum(kube_persistentvolumeclaim_info{namespace=~"$1"})`,
|
||||
"workspace_daemonset_count": `sum(kube_daemonset_labels{namespace=~"$1"})`,
|
||||
"workspace_deployment_count": `sum(kube_deployment_labels{namespace=~"$1"})`,
|
||||
"workspace_endpoint_count": `sum(kube_endpoint_labels{namespace=~"$1"})`,
|
||||
"workspace_hpa_count": `sum(kube_hpa_labels{namespace=~"$1"})`,
|
||||
"workspace_job_count": `sum(kube_job_labels{namespace=~"$1"})`,
|
||||
"workspace_statefulset_count": `sum(kube_statefulset_labels{namespace=~"$1"})`,
|
||||
"workspace_replicaset_count": `count(kube_replicaset_created{namespace=~"$1"})`,
|
||||
"workspace_service_count": `sum(kube_service_info{namespace=~"$1"})`,
|
||||
"workspace_secret_count": `sum(kube_secret_info{namespace=~"$1"})`,
|
||||
"workspace_cronjob_count": `sum(kube_cronjob_labels{namespace!="", namespace$1})`,
|
||||
"workspace_pvc_count": `sum(kube_persistentvolumeclaim_info{namespace!="", namespace$1})`,
|
||||
"workspace_daemonset_count": `sum(kube_daemonset_labels{namespace!="", namespace$1})`,
|
||||
"workspace_deployment_count": `sum(kube_deployment_labels{namespace!="", namespace$1})`,
|
||||
"workspace_endpoint_count": `sum(kube_endpoint_labels{namespace!="", namespace$1})`,
|
||||
"workspace_hpa_count": `sum(kube_hpa_labels{namespace!="", namespace$1})`,
|
||||
"workspace_job_count": `sum(kube_job_labels{namespace!="", namespace$1})`,
|
||||
"workspace_statefulset_count": `sum(kube_statefulset_labels{namespace!="", namespace$1})`,
|
||||
"workspace_replicaset_count": `count(kube_replicaset_created{namespace!="", namespace$1})`,
|
||||
"workspace_service_count": `sum(kube_service_info{namespace!="", namespace$1})`,
|
||||
"workspace_secret_count": `sum(kube_secret_info{namespace!="", namespace$1})`,
|
||||
|
||||
"workspace_all_project_count": `count(kube_namespace_annotations)`,
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user