project restructuring and initial refactoring

main/prometheus/helm.ts

@@ -0,0 +1,29 @@
+import { PrometheusLens } from "./lens"
+import { CoreV1Api } from "@kubernetes/client-node"
+import { PrometheusService } from "./provider-registry";
+import logger from "../logger"
+
+export class PrometheusHelm extends PrometheusLens {
+  id = "helm"
+  name = "Helm"
+  rateAccuracy = "5m"
+
+  public async getPrometheusService(client: CoreV1Api): Promise<PrometheusService> {
+    const labelSelector = "app=prometheus,component=server,heritage=Helm"
+    try {
+      const serviceList = await client.listServiceForAllNamespaces(false, "", null, labelSelector)
+      const service = serviceList.body.items[0]
+      if (!service) return
+
+      return {
+        id: this.id,
+        namespace: service.metadata.namespace,
+        service: service.metadata.name,
+        port: service.spec.ports[0].port
+      }
+    } catch(error) {
+      logger.warn(`PrometheusHelm: failed to list services: ${error.toString()}`)
+      return
+    }
+  }
+}

main/prometheus/lens.ts

@@ -0,0 +1,83 @@
+import { PrometheusProvider, PrometheusQueryOpts, PrometheusQuery, PrometheusService } from "./provider-registry";
+import { CoreV1Api } from "@kubernetes/client-node";
+import logger from "../logger"
+
+export class PrometheusLens implements PrometheusProvider {
+  id = "lens"
+  name = "Lens"
+  rateAccuracy = "1m"
+
+  public async getPrometheusService(client: CoreV1Api): Promise<PrometheusService> {
+    try {
+      const resp = await client.readNamespacedService("prometheus", "lens-metrics")
+      const service = resp.body
+      return {
+        id: this.id,
+        namespace: service.metadata.namespace,
+        service: service.metadata.name,
+        port: service.spec.ports[0].port
+      }
+    } catch(error) {
+      logger.warn(`PrometheusLens: failed to list services: ${error.toString()}`)
+    }
+  }
+
+  public getQueries(opts: PrometheusQueryOpts): PrometheusQuery {
+    switch(opts.category) {
+    case 'cluster':
+      return {
+        memoryUsage: `
+          sum(
+            node_memory_MemTotal_bytes - (node_memory_MemFree_bytes + node_memory_Buffers_bytes + node_memory_Cached_bytes)
+          ) by (kubernetes_name)
+        `.replace(/_bytes/g, `_bytes{kubernetes_node=~"${opts.nodes}"}`),
+        memoryRequests: `sum(kube_pod_container_resource_requests{node=~"${opts.nodes}", resource="memory"}) by (component)`,
+        memoryLimits: `sum(kube_pod_container_resource_limits{node=~"${opts.nodes}", resource="memory"}) by (component)`,
+        memoryCapacity: `sum(kube_node_status_capacity{node=~"${opts.nodes}", resource="memory"}) by (component)`,
+        cpuUsage: `sum(rate(node_cpu_seconds_total{kubernetes_node=~"${opts.nodes}", mode=~"user|system"}[${this.rateAccuracy}]))`,
+        cpuRequests:`sum(kube_pod_container_resource_requests{node=~"${opts.nodes}", resource="cpu"}) by (component)`,
+        cpuLimits: `sum(kube_pod_container_resource_limits{node=~"${opts.nodes}", resource="cpu"}) by (component)`,
+        cpuCapacity: `sum(kube_node_status_capacity{node=~"${opts.nodes}", resource="cpu"}) by (component)`,
+        podUsage: `sum(kubelet_running_pod_count{instance=~"${opts.nodes}"})`,
+        podCapacity: `sum(kube_node_status_capacity{node=~"${opts.nodes}", resource="pods"}) by (component)`,
+        fsSize: `sum(node_filesystem_size_bytes{kubernetes_node=~"${opts.nodes}", mountpoint="/"}) by (kubernetes_node)`,
+        fsUsage: `sum(node_filesystem_size_bytes{kubernetes_node=~"${opts.nodes}", mountpoint="/"} - node_filesystem_avail_bytes{kubernetes_node=~"${opts.nodes}", mountpoint="/"}) by (kubernetes_node)`
+      }
+    case 'nodes':
+      return {
+        memoryUsage: `sum (node_memory_MemTotal_bytes - (node_memory_MemFree_bytes + node_memory_Buffers_bytes + node_memory_Cached_bytes)) by (kubernetes_node)`,
+        memoryCapacity: `sum(kube_node_status_capacity{resource="memory"}) by (node)`,
+        cpuUsage: `sum(rate(node_cpu_seconds_total{mode=~"user|system"}[${this.rateAccuracy}])) by(kubernetes_node)`,
+        cpuCapacity: `sum(kube_node_status_allocatable{resource="cpu"}) by (node)`,
+        fsSize: `sum(node_filesystem_size_bytes{mountpoint="/"}) by (kubernetes_node)`,
+        fsUsage: `sum(node_filesystem_size_bytes{mountpoint="/"} - node_filesystem_avail_bytes{mountpoint="/"}) by (kubernetes_node)`
+      }
+    case 'pods':
+      return {
+        cpuUsage: `sum(rate(container_cpu_usage_seconds_total{container!="POD",container!="",pod=~"${opts.pods}",namespace="${opts.namespace}"}[${this.rateAccuracy}])) by (${opts.selector})`,
+        cpuRequests: `sum(kube_pod_container_resource_requests{pod=~"${opts.pods}",resource="cpu",namespace="${opts.namespace}"}) by (${opts.selector})`,
+        cpuLimits: `sum(kube_pod_container_resource_limits{pod=~"${opts.pods}",resource="cpu",namespace="${opts.namespace}"}) by (${opts.selector})`,
+        memoryUsage: `sum(container_memory_working_set_bytes{container!="POD",container!="",pod=~"${opts.pods}",namespace="${opts.namespace}"}) by (${opts.selector})`,
+        memoryRequests: `sum(kube_pod_container_resource_requests{pod=~"${opts.pods}",resource="memory",namespace="${opts.namespace}"}) by (${opts.selector})`,
+        memoryLimits: `sum(kube_pod_container_resource_limits{pod=~"${opts.pods}",resource="memory",namespace="${opts.namespace}"}) by (${opts.selector})`,
+        fsUsage: `sum(container_fs_usage_bytes{container!="POD",container!="",pod=~"${opts.pods}",namespace="${opts.namespace}"}) by (${opts.selector})`,
+        networkReceive: `sum(rate(container_network_receive_bytes_total{pod=~"${opts.pods}",namespace="${opts.namespace}"}[${this.rateAccuracy}])) by (${opts.selector})`,
+        networkTransmit: `sum(rate(container_network_transmit_bytes_total{pod=~"${opts.pods}",namespace="${opts.namespace}"}[${this.rateAccuracy}])) by (${opts.selector})`
+      }
+    case 'pvc':
+      return {
+        diskUsage: `sum(kubelet_volume_stats_used_bytes{persistentvolumeclaim="${opts.pvc}"}) by (persistentvolumeclaim, namespace)`,
+        diskCapacity: `sum(kubelet_volume_stats_capacity_bytes{persistentvolumeclaim="${opts.pvc}"}) by (persistentvolumeclaim, namespace)`
+      }
+    case 'ingress':
+      const bytesSent = (ingress: string, statuses: string) =>
+        `sum(rate(nginx_ingress_controller_bytes_sent_sum{ingress="${ingress}", status=~"${statuses}"}[${this.rateAccuracy}])) by (ingress)`
+      return {
+        bytesSentSuccess: bytesSent(opts.igress, "^2\\\\d*"),
+        bytesSentFailure: bytesSent(opts.ingres, "^5\\\\d*"),
+        requestDurationSeconds: `sum(rate(nginx_ingress_controller_request_duration_seconds_sum{ingress="${opts.ingress}"}[${this.rateAccuracy}])) by (ingress)`,
+        responseDurationSeconds: `sum(rate(nginx_ingress_controller_response_duration_seconds_sum{ingress="${opts.ingress}"}[${this.rateAccuracy}])) by (ingress)`
+      }
+    }
+  }
+}

main/prometheus/operator.ts

@@ -0,0 +1,91 @@
+import { PrometheusProvider, PrometheusQueryOpts, PrometheusQuery, PrometheusService } from "./provider-registry";
+import { CoreV1Api, V1Service } from "@kubernetes/client-node";
+import logger from "../logger";
+
+export class PrometheusOperator implements PrometheusProvider {
+  rateAccuracy = "1m"
+  id = "operator"
+  name = "Prometheus Operator"
+
+  public async getPrometheusService(client: CoreV1Api): Promise<PrometheusService> {
+    try {
+      let service: V1Service
+      for (const labelSelector of ["operated-prometheus=true", "self-monitor=true"]) {
+        if (!service) {
+          const serviceList = await client.listServiceForAllNamespaces(null, null, null, labelSelector)
+          service = serviceList.body.items[0]
+        }
+      }
+      if (!service) return
+
+      return {
+        id: this.id,
+        namespace: service.metadata.namespace,
+        service: service.metadata.name,
+        port: service.spec.ports[0].port
+      }
+    } catch(error) {
+      logger.warn(`PrometheusOperator: failed to list services: ${error.toString()}`)
+      return
+    }
+  }
+
+  public getQueries(opts: PrometheusQueryOpts): PrometheusQuery {
+    switch(opts.category) {
+    case 'cluster':
+      return {
+        memoryUsage: `
+          sum(
+            node_memory_MemTotal_bytes - (node_memory_MemFree_bytes + node_memory_Buffers_bytes + node_memory_Cached_bytes)
+          ) by (node)
+        `.replace(/_bytes/g, `_bytes * on (pod,namespace) group_left(node) kube_pod_info{node=~"${opts.nodes}"}`),
+        memoryRequests: `sum(kube_pod_container_resource_requests{node=~"${opts.nodes}", resource="memory"}) by (component)`,
+        memoryLimits: `sum(kube_pod_container_resource_limits{node=~"${opts.nodes}", resource="memory"}) by (component)`,
+        memoryCapacity: `sum(kube_node_status_capacity{node=~"${opts.nodes}", resource="memory"}) by (component)`,
+        cpuUsage: `sum(rate(node_cpu_seconds_total{mode=~"user|system"}[${this.rateAccuracy}])* on (pod,namespace) group_left(node) kube_pod_info{node=~"${opts.nodes}"}) by (node)`,
+        cpuRequests:`sum(kube_pod_container_resource_requests{node=~"${opts.nodes}", resource="cpu"}) by (component)`,
+        cpuLimits: `sum(kube_pod_container_resource_limits{node=~"${opts.nodes}", resource="cpu"}) by (component)`,
+        cpuCapacity: `sum(kube_node_status_capacity{node=~"${opts.nodes}", resource="cpu"}) by (component)`,
+        podUsage: `sum(kubelet_running_pod_count{node=~"${opts.nodes}"})`,
+        podCapacity: `sum(kube_node_status_capacity{node=~"${opts.nodes}", resource="pods"}) by (component)`,
+        fsSize: `sum(node_filesystem_size_bytes{mountpoint="/"} * on (pod,namespace) group_left(node) kube_pod_info{node=~"${opts.nodes}"}) by (node)`,
+        fsUsage: `sum(node_filesystem_size_bytes{mountpoint="/"} * on (pod,namespace) group_left(node) kube_pod_info{node=~"${opts.nodes}"} - node_filesystem_avail_bytes{mountpoint="/"} * on (pod,namespace) group_left(node) kube_pod_info{node=~"${opts.nodes}"}) by (node)`
+      }
+    case 'nodes':
+      return {
+        memoryUsage: `sum((node_memory_MemTotal_bytes - (node_memory_MemFree_bytes + node_memory_Buffers_bytes + node_memory_Cached_bytes)) * on (pod,namespace) group_left(node) kube_pod_info) by (node)`,
+        memoryCapacity: `sum(kube_node_status_capacity{resource="memory"}) by (node)`,
+        cpuUsage: `sum(rate(node_cpu_seconds_total{mode=~"user|system"}[${this.rateAccuracy}]) * on (pod,namespace) group_left(node) kube_pod_info) by (node)`,
+        cpuCapacity: `sum(kube_node_status_allocatable{resource="cpu"}) by (node)`,
+        fsSize: `sum(node_filesystem_size_bytes{mountpoint="/"} * on (pod,namespace) group_left(node) kube_pod_info) by (node)`,
+        fsUsage: `sum((node_filesystem_size_bytes{mountpoint="/"} - node_filesystem_avail_bytes{mountpoint="/"}) * on (pod,namespace) group_left(node) kube_pod_info) by (node)`
+      }
+    case 'pods':
+      return {
+        cpuUsage: `sum(rate(container_cpu_usage_seconds_total{container!="POD",container!="",pod=~"${opts.pods}",namespace="${opts.namespace}"}[${this.rateAccuracy}])) by (${opts.selector})`,
+        cpuRequests: `sum(kube_pod_container_resource_requests{pod=~"${opts.pods}",resource="cpu",namespace="${opts.namespace}"}) by (${opts.selector})`,
+        cpuLimits: `sum(kube_pod_container_resource_limits{pod=~"${opts.pods}",resource="cpu",namespace="${opts.namespace}"}) by (${opts.selector})`,
+        memoryUsage: `sum(container_memory_working_set_bytes{container!="POD",container!="",pod=~"${opts.pods}",namespace="${opts.namespace}"}) by (${opts.selector})`,
+        memoryRequests: `sum(kube_pod_container_resource_requests{pod=~"${opts.pods}",resource="memory",namespace="${opts.namespace}"}) by (${opts.selector})`,
+        memoryLimits: `sum(kube_pod_container_resource_limits{pod=~"${opts.pods}",resource="memory",namespace="${opts.namespace}"}) by (${opts.selector})`,
+        fsUsage: `sum(container_fs_usage_bytes{container!="POD",container!="",pod=~"${opts.pods}",namespace="${opts.namespace}"}) by (${opts.selector})`,
+        networkReceive: `sum(rate(container_network_receive_bytes_total{pod=~"${opts.pods}",namespace="${opts.namespace}"}[${this.rateAccuracy}])) by (${opts.selector})`,
+        networkTransmit: `sum(rate(container_network_transmit_bytes_total{pod=~"${opts.pods}",namespace="${opts.namespace}"}[${this.rateAccuracy}])) by (${opts.selector})`
+      }
+    case 'pvc':
+      return {
+        diskUsage: `sum(kubelet_volume_stats_used_bytes{persistentvolumeclaim="${opts.pvc}"}) by (persistentvolumeclaim, namespace)`,
+        diskCapacity: `sum(kubelet_volume_stats_capacity_bytes{persistentvolumeclaim="${opts.pvc}"}) by (persistentvolumeclaim, namespace)`
+      }
+    case 'ingress':
+      const bytesSent = (ingress: string, statuses: string) =>
+        `sum(rate(nginx_ingress_controller_bytes_sent_sum{ingress="${ingress}", status=~"${statuses}"}[${this.rateAccuracy}])) by (ingress)`;
+      return {
+        bytesSentSuccess: bytesSent(opts.igress, "^2\\\\d*"),
+        bytesSentFailure: bytesSent(opts.ingres, "^5\\\\d*"),
+        requestDurationSeconds: `sum(rate(nginx_ingress_controller_request_duration_seconds_sum{ingress="${opts.ingress}"}[${this.rateAccuracy}])) by (ingress)`,
+        responseDurationSeconds: `sum(rate(nginx_ingress_controller_response_duration_seconds_sum{ingress="${opts.ingress}"}[${this.rateAccuracy}])) by (ingress)`
+      }
+    }
+  }
+}

main/prometheus/provider-registry.ts

@@ -0,0 +1,90 @@
+import { CoreV1Api } from "@kubernetes/client-node"
+
+export type PrometheusClusterQuery = {
+  memoryUsage: string;
+  memoryRequests: string;
+  memoryLimits: string;
+  memoryCapacity: string;
+  cpuUsage: string;
+  cpuRequests: string;
+  cpuLimits: string;
+  cpuCapacity: string;
+  podUsage: string;
+  podCapacity: string;
+}
+
+export type PrometheusNodeQuery = {
+  memoryUsage: string;
+  memoryCapacity: string;
+  cpuUsage: string;
+  cpuCapacity: string;
+  fsSize: string;
+  fsUsage: string;
+}
+
+export type PrometheusPodQuery = {
+  memoryUsage: string;
+  memoryRequests: string;
+  memoryLimits: string;
+  cpuUsage: string;
+  cpuRequests: string;
+  cpuLimits: string;
+  fsUsage: string;
+  networkReceive: string;
+  networkTransmit: string;
+}
+
+export type PrometheusPvcQuery = {
+  diskUsage: string;
+  diskCapacity: string;
+}
+
+export type PrometheusIngressQuery = {
+  bytesSentSuccess: string;
+  bytesSentFailure: string;
+  requestDurationSeconds: string;
+  responseDurationSeconds: string;
+}
+
+export type PrometheusQueryOpts = {
+  [key: string]: string | any;
+};
+
+export type PrometheusQuery = PrometheusNodeQuery | PrometheusClusterQuery | PrometheusPodQuery | PrometheusPvcQuery | PrometheusIngressQuery
+
+export type PrometheusService = {
+  id: string;
+  namespace: string;
+  service: string;
+  port: number;
+}
+
+export interface PrometheusProvider {
+  id: string;
+  name: string;
+  getQueries(opts: PrometheusQueryOpts): PrometheusQuery;
+  getPrometheusService(client: CoreV1Api): Promise<PrometheusService>;
+}
+
+export type PrometheusProviderList = {
+  [key: string]: PrometheusProvider;
+}
+
+export class PrometheusProviderRegistry {
+  private static prometheusProviders: PrometheusProviderList = {}
+
+  static getProvider(type: string): PrometheusProvider {
+    if (!this.prometheusProviders[type]) {
+      throw "Unknown Prometheus provider";
+    }
+    return this.prometheusProviders[type]
+  }
+
+  static registerProvider(key: string, provider: PrometheusProvider) {
+    this.prometheusProviders[key] = provider
+  }
+
+  static getProviders(): PrometheusProvider[] {
+    return Object.values(this.prometheusProviders)
+  }
+}