Pods 之间 GKE 中的不良延迟

Question

我们有一个非常奇怪的行为，在 kubernetes 集群 (GKE) 内通信有不可接受的大延迟。对于具有 Memorystore get/store 操作和 CloudSQL 查询的端点，延迟在 600 毫秒和 1 秒之间跳跃。在开发环境中本地相同的设置运行（尽管没有 k8s）没有显示这种延迟。

关于我们的架构：我们是运行 GKE 上的一个 k8s 集群，使用 terraform 和服务/部署 (yaml) 文件进行创建（我在下面添加了这些）。我们运行两个节点 APIs (koa.js 2.5)。一个 API 暴露在 public 的入口，并通过节点端口连接到 API pod。

另一个 API pod 可通过内部负载均衡器从 google 私有访问。此 API 已连接到我们需要的所有资源（CloudSQL、Cloud Storage）。

两个 APIs 也连接到 Memorystore (Redis)。

pods 之间的通信使用自签名 server/client 证书进行保护（这不是问题，我们已经暂时将其删除以进行测试）。

我们检查了日志，发现从 public API 到私人请求只需要大约 200 毫秒就可以到达。此外，私有 public API 的响应花费了大约 600 毫秒（从私有 API 的整个业务逻辑抛出到我们在公开 API)

我们真的没有什么可以尝试的了......我们已经将所有 Google 云资源连接到我们的本地环境，这并没有显示出那种糟糕的延迟。

在完整的本地设置中，延迟仅为我们在云设置中看到的延迟的 1/5 到 1/10。我们还尝试从位于 0.100 毫秒区域的 public ping 私有 POD。

您有什么想法可以让我们进一步调查吗？这是关于我们的 Google 云设置

的地形脚本

  // Configure the Google Cloud provider
  provider "google" {
    project = "${var.project}"
    region  = "${var.region}"
  }
  data "google_compute_zones" "available" {}
  # Ensuring relevant service APIs are enabled in your project. Alternatively visit and enable the needed services
  resource "google_project_service" "serviceapi" {
    service            = "serviceusage.googleapis.com"
    disable_on_destroy = false
  }
  resource "google_project_service" "sqlapi" {
    service            = "sqladmin.googleapis.com"
    disable_on_destroy = false
    depends_on         = ["google_project_service.serviceapi"]
  }
  resource "google_project_service" "redisapi" {
    service            = "redis.googleapis.com"
    disable_on_destroy = false
    depends_on         = ["google_project_service.serviceapi"]
  }
  # Create a VPC and a subnetwork in our region
  resource "google_compute_network" "appnetwork" {
    name                    = "${var.environment}-vpn"
    auto_create_subnetworks = "false"
  }
  resource "google_compute_subnetwork" "network-with-private-secondary-ip-ranges" {
    name          = "${var.environment}-vpn-subnet"
    ip_cidr_range = "10.2.0.0/16"
    region        = "europe-west1"
    network       = "${google_compute_network.appnetwork.self_link}"
    secondary_ip_range {
      range_name    = "kubernetes-secondary-range-pods"
      ip_cidr_range = "10.60.0.0/16"
    }
    secondary_ip_range {
      range_name    = "kubernetes-secondary-range-services"
      ip_cidr_range = "10.70.0.0/16"
    }
  }
  # GKE cluster setup
  resource "google_container_cluster" "primary" {
    name               = "${var.environment}-cluster"
    zone               = "${data.google_compute_zones.available.names[1]}"
    initial_node_count = 1
    description        = "Kubernetes Cluster"
    network            = "${google_compute_network.appnetwork.self_link}"
    subnetwork         = "${google_compute_subnetwork.network-with-private-secondary-ip-ranges.self_link}"
    depends_on         = ["google_project_service.serviceapi"]
    additional_zones = [
      "${data.google_compute_zones.available.names[0]}",
      "${data.google_compute_zones.available.names[2]}",
    ]
    master_auth {
      username = "xxxxxxx"
      password = "xxxxxxx"
    }
    ip_allocation_policy {
      cluster_secondary_range_name  = "kubernetes-secondary-range-pods"
      services_secondary_range_name = "kubernetes-secondary-range-services"
    }
    node_config {
      oauth_scopes = [
        "https://www.googleapis.com/auth/compute",
        "https://www.googleapis.com/auth/devstorage.read_only",
        "https://www.googleapis.com/auth/logging.write",
        "https://www.googleapis.com/auth/monitoring",
        "https://www.googleapis.com/auth/trace.append"
      ]
      tags = ["kubernetes", "${var.environment}"]
    }
  }
  ##################
  # MySQL DATABASES 
  ##################
  resource "google_sql_database_instance" "core" {
    name             = "${var.environment}-sql-core"
    database_version = "MYSQL_5_7"
    region           = "${var.region}"
    depends_on       = ["google_project_service.sqlapi"]
    settings {
      # Second-generation instance tiers are based on the machine
      # type. See argument reference below.
      tier = "db-n1-standard-1"
    }
  }
  resource "google_sql_database_instance" "tenant1" {
    name             = "${var.environment}-sql-tenant1"
    database_version = "MYSQL_5_7"
    region           = "${var.region}"
    depends_on       = ["google_project_service.sqlapi"]
    settings {
      # Second-generation instance tiers are based on the machine
      # type. See argument reference below.
      tier = "db-n1-standard-1"
    }
  }
  resource "google_sql_database_instance" "tenant2" {
    name             = "${var.environment}-sql-tenant2"
    database_version = "MYSQL_5_7"
    region           = "${var.region}"
    depends_on       = ["google_project_service.sqlapi"]
    settings {
      # Second-generation instance tiers are based on the machine
      # type. See argument reference below.
      tier = "db-n1-standard-1"
    }
  }
  resource "google_sql_database" "core" {
    name     = "project_core"
    instance = "${google_sql_database_instance.core.name}"
  }
  resource "google_sql_database" "tenant1" {
    name     = "project_tenant_1"
    instance = "${google_sql_database_instance.tenant1.name}"
  }
  resource "google_sql_database" "tenant2" {
    name     = "project_tenant_2"
    instance = "${google_sql_database_instance.tenant2.name}"
  }
  ##################
  # MySQL USERS
  ##################
  resource "google_sql_user" "core-user" {
    name     = "${var.sqluser}"
    instance = "${google_sql_database_instance.core.name}"
    host     = "cloudsqlproxy~%"
    password = "${var.sqlpassword}"
  }
  resource "google_sql_user" "tenant1-user" {
    name     = "${var.sqluser}"
    instance = "${google_sql_database_instance.tenant1.name}"
    host     = "cloudsqlproxy~%"
    password = "${var.sqlpassword}"
  }
  resource "google_sql_user" "tenant2-user" {
    name     = "${var.sqluser}"
    instance = "${google_sql_database_instance.tenant2.name}"
    host     = "cloudsqlproxy~%"
    password = "${var.sqlpassword}"
  }
  ##################
  # REDIS
  ##################
  resource "google_redis_instance" "redis" {
    name               = "${var.environment}-redis"
    tier               = "BASIC"
    memory_size_gb     = 1
    depends_on         = ["google_project_service.redisapi"]
    authorized_network = "${google_compute_network.appnetwork.self_link}"
    region             = "${var.region}"
    location_id        = "${data.google_compute_zones.available.names[0]}"
    redis_version = "REDIS_3_2"
    display_name  = "Redis Instance"
  }
  # The following outputs allow authentication and connectivity to the GKE Cluster.
  output "client_certificate" {
    value = "${google_container_cluster.primary.master_auth.0.client_certificate}"
  }
  output "client_key" {
    value = "${google_container_cluster.primary.master_auth.0.client_key}"
  }
  output "cluster_ca_certificate" {
    value = "${google_container_cluster.primary.master_auth.0.cluster_ca_certificate}"
  }

私有的服务和部署API

  # START CRUD POD
  apiVersion: extensions/v1beta1
  kind: Deployment
  metadata:
    name: crud-pod
    labels:
      app: crud
  spec:
    template:
      metadata:
        labels:
          app: crud
      spec:
        containers:
          - name: crud
            image: eu.gcr.io/dev-xxxxx/crud:latest-unstable
            ports:
              - containerPort: 3333
            env:
            - name: NODE_ENV
              value: develop
            volumeMounts:
            - [..MountedConfigFiles..]
          # [START proxy_container]
          - name: cloudsql-proxy
            image: gcr.io/cloudsql-docker/gce-proxy:1.11
            command: ["/cloud_sql_proxy",
                      "-instances=dev-xxxx:europe-west1:dev-sql-core=tcp:3306,dev-xxxx:europe-west1:dev-sql-tenant1=tcp:3307,dev-xxxx:europe-west1:dev-sql-tenant2=tcp:3308",
                      "-credential_file=xxxx"]
            volumeMounts:
              - name: cloudsql-instance-credentials
                mountPath: /secrets/cloudsql
                readOnly: true
          # [END proxy_container]
        # [START volumes]
        volumes:
          - name: cloudsql-instance-credentials
            secret:
              secretName: cloudsql-instance-credentials
          - [..ConfigFilesVolumes..]
        # [END volumes]
  # END CRUD POD
  -------
  # START CRUD SERVICE
  apiVersion: v1
  kind: Service
  metadata:
    name: crud
    annotations:
      cloud.google.com/load-balancer-type: "Internal"
  spec:
    type: LoadBalancer
    loadBalancerSourceRanges: 
      - 10.60.0.0/16
    ports:
    - name: crud-port
      port: 3333
      protocol: TCP # default; can also specify UDP
    selector:
      app: crud # label selector for Pods to target
  # END CRUD SERVICE

还有 public 个（包括入口）

  # START SAPI POD
  apiVersion: extensions/v1beta1
  kind: Deployment
  metadata:
    name: sapi-pod
    labels:
      app: sapi
  spec:
    template:
      metadata:
        labels:
          app: sapi
      spec:
        containers:
          - name: sapi
            image: eu.gcr.io/dev-xxx/sapi:latest-unstable
            ports:
              - containerPort: 8080
            env:
              - name: NODE_ENV
                value: develop
            volumeMounts:
              - [..MountedConfigFiles..]
        volumes:
          - [..ConfigFilesVolumes..]
  # END SAPI POD
  -------------
  # START SAPI SERVICE
  kind: Service
  apiVersion: v1
  metadata:
    name: sapi # Service name
  spec:
    selector:
      app:  sapi
    ports:
    - port: 8080
      targetPort: 8080
    type: NodePort
  # END SAPI SERVICE
  --------------
  apiVersion: extensions/v1beta1
  kind: Ingress
  metadata:
    name: dev-ingress
    annotations:
      kubernetes.io/ingress.global-static-ip-name: api-dev-static-ip
    labels:
      app: sapi-ingress
  spec:
    backend:
      serviceName: sapi
      servicePort: 8080
    tls:
    - hosts:
      - xxxxx
      secretName: xxxxx

Answer 1

我们通过从 logTransport 中删除 @google-cloud/logging-winston 解决了这个问题。出于某种原因，它阻塞了我们的流量，导致我们的延迟非常糟糕。

Pods 之间 GKE 中的不良延迟

Bad latency in GKE between Pods

rest

google-cloud-platform

kubernetes

google-kubernetes-engine

internal-load-balancer