iT邦幫忙

2022 iThome 鐵人賽
DAY 27
0
DevOps

那些文件沒告訴你的 AWS EKS系列 第 27 篇

[27] 為什麼 EKS 在更新 Kubernetes deployment 時會有 HTTP 502 的 error（二）

14th鐵人賽 eks kubernetes
1289 瀏覽

  • 分享至 

  • xImage
    •  

接續前一天，我們透過 AWS Load Balancer Controller 作為 Ingress Controller 部署了 Application Load Balancers（ALB）但是在更新 deployment 時卻產生了 HTTP 502 的 error。

一般來說，ALB 響應 HTTP 502 [1]原因來自於 Load Balancer 嘗試建立連線時收到 target 的 TCP RST。根據測試步驟來說，我們在嘗試更新 deployment image，此動作對於 Kubernetes 可分為兩個部分：

  1. Service 更新：每一個 Service object 上定義了相應 Service Port 及 Deployment object，實際 Kubernetes 透過 selector 選取後端 Pod，這一步抽象化儲存為 Endpoints object。因此在 Deployment rolling update 時，Pod 更新意味著 Endpoints 也須即時更新 IP 地址資訊。以下為查看 Endpoints 及 Service object 輸出，可以觀察到 Service 上 Endpoints 與 Endpoints 輸出一致：
$ kubectl -n demo-nginx describe svc,endpoints
Name:              service-demo-nginx
Namespace:         demo-nginx
Labels:            <none>
Annotations:       <none>
Selector:          app=demo-nginx
Type:              ClusterIP
IP Family Policy:  SingleStack
IP Families:       IPv4
IP:                10.100.247.243
IPs:               10.100.247.243
Port:              <unset>  80/TCP
TargetPort:        80/TCP
Endpoints:         192.168.16.64:80,192.168.31.134:80,192.168.40.177:80 + 7 more...
Session Affinity:  None
Events:            <none>

Name:         service-demo-nginx
Namespace:    demo-nginx
Labels:       <none>
Annotations:  endpoints.kubernetes.io/last-change-trigger-time: 2022-10-11T15:39:57Z
Subsets:
  Addresses:          192.168.16.64,192.168.31.134,192.168.40.177,192.168.41.36,192.168.45.23,192.168.48.198,192.168.64.124,192.168.66.152,192.168.67.240,192.168.91.189
  NotReadyAddresses:  <none>
  Ports:
    Name     Port  Protocol
    ----     ----  --------
    <unset>  80    TCP

Events:  <none>

透過 CloudWatch Log Insight syntax 查看 Endpoints 更新：

filter @logStream like /^kube-apiserver-audit/
 | fields requestReceivedTimestamp
 | sort @timestamp asc
 | filter objectRef.resource == 'endpoints' AND objectRef.name == 'service-demo-nginx' AND verb == 'update'
 | limit 10000

從輸出可以觀察到將近數十筆關於 Endpoints 更新 subnets IP 地址，正也是 Deployment rolling upgrade 時 Pod IP 更新。以下擷取一筆部分紀錄：

    "@message": {
      "kind": "Event",
      "apiVersion": "audit.k8s.io/v1",
      "level": "RequestResponse",
      "auditID": "7db6a466-c7ff-45a2-b3e8-9d21de586c3b",
      "stage": "ResponseComplete",
      "requestURI": "/api/v1/namespaces/demo-nginx/endpoints/service-demo-nginx",
      "verb": "update",
      "user": {
        "username": "system:serviceaccount:kube-system:endpoint-controller",
        ...
        ...
      "requestObject": {
        "kind": "Endpoints",
        "apiVersion": "v1",
        "metadata": {
          "name": "service-demo-nginx",
          "namespace": "demo-nginx",
          "uid": "2468c3ba-1bb3-4740-957b-b7a0ce58f9d0",
          "resourceVersion": "7959165",
          "creationTimestamp": "2022-10-11T14:31:21Z",
      ...
      ...
              "subsets": [
          {
            "addresses": [
              {
                "ip": "192.168.19.10",
                "nodeName": "ip-192-168-18-190.eu-west-1.compute.internal",
                "targetRef": {
                  "kind": "Pod",
                  "namespace": "demo-nginx",
                  "name": "demo-nginx-deployment-78dc5d6bdb-gdtwz",
                  "uid": "bf69b974-c06f-49b9-a09f-308e76aabe11",
                  "resourceVersion": "7958990"
                }
              },
              {
                "ip": "192.168.35.103",
                "nodeName": "ip-192-168-55-245.eu-west-1.compute.internal",
                "targetRef": {
                  "kind": "Pod",
                  "namespace": "demo-nginx",
                  "name": "demo-nginx-deployment-78dc5d6bdb-9xctc",
                  "uid": "eada6e94-6dc9-4623-b796-5dd720402ab7",
                  "resourceVersion": "7959042"
                }
              },
      ...
      ...
      ...
  1. Deployment 更新：更新 Deployment Object 內 image。
  2. Ingress Controller 更新：以 AWS Load Balancer Controller 負責基於 Ingress YAML 設定 ALB 及 Target Group。如同 Service object 狀態一樣，當 Pod IP 更新時 Ingress controller 也須即時更新（registering/deregistering）ALB 資源上 Target group。

因此也能可以於 AWS Load Balancer Controller Pod 上過濾 Target Group id 查看到 registering/deregistering 更新對應 IP 地址。

$ kubectl -n kube-system logs aws-load-balancer-controller-b9b6478cf-ztfrd | grep "k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2"
...
...
...
{"level":"info","ts":1665502791.4772518,"msg":"deRegistering targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2","targets":[{"AvailabilityZone":"eu-west-1a","Id":"192.168.22.141","Port":80},{"AvailabilityZone":"eu-west-1b","Id":"192.168.88.25","Port":80}]}
{"level":"info","ts":1665502791.5222945,"msg":"deRegistered targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2"}
{"level":"info","ts":1665502793.781365,"msg":"registering targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2","targets":[{"AvailabilityZone":null,"Id":"192.168.45.23","Port":80}]}
{"level":"info","ts":1665502794.109826,"msg":"registered targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2"}
{"level":"info","ts":1665502794.2468624,"msg":"deRegistering targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2","targets":[{"AvailabilityZone":"eu-west-1c","Id":"192.168.38.205","Port":80},{"AvailabilityZone":"eu-west-1c","Id":"192.168.53.217","Port":80}]}
{"level":"info","ts":1665502794.2834873,"msg":"deRegistered targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2"}
{"level":"info","ts":1665502794.2835267,"msg":"registering targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2","targets":[{"AvailabilityZone":null,"Id":"192.168.40.177","Port":80}]}
{"level":"info","ts":1665502794.4290822,"msg":"registered targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2"}
{"level":"info","ts":1665502794.5652397,"msg":"deRegistering targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2","targets":[{"AvailabilityZone":"eu-west-1b","Id":"192.168.76.79","Port":80},{"AvailabilityZone":"eu-west-1b","Id":"192.168.90.183","Port":80}]}
{"level":"info","ts":1665502794.6063979,"msg":"deRegistered targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2"}
{"level":"info","ts":1665502794.6082087,"msg":"registering targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2","targets":[{"AvailabilityZone":null,"Id":"192.168.31.134","Port":80},{"AvailabilityZone":null,"Id":"192.168.67.240","Port":80}]}
{"level":"info","ts":1665502794.7579286,"msg":"registered targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2"}
{"level":"info","ts":1665502795.0438414,"msg":"deRegistering targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2","targets":[{"AvailabilityZone":"eu-west-1a","Id":"192.168.19.10","Port":80}]}
{"level":"info","ts":1665502795.0870168,"msg":"deRegistered targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2"}
{"level":"info","ts":1665502795.0870724,"msg":"registering targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2","targets":[{"AvailabilityZone":null,"Id":"192.168.66.152","Port":80}]}
{"level":"info","ts":1665502795.2049546,"msg":"registered targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2"}
{"level":"info","ts":1665502796.5167701,"msg":"registering targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2","targets":[{"AvailabilityZone":null,"Id":"192.168.91.189","Port":80}]}
{"level":"info","ts":1665502796.6328108,"msg":"registered targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2"}
{"level":"info","ts":1665502796.765318,"msg":"deRegistering targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2","targets":[{"AvailabilityZone":"eu-west-1c","Id":"192.168.35.103","Port":80},{"AvailabilityZone":"eu-west-1c","Id":"192.168.61.232","Port":80}]}
{"level":"info","ts":1665502796.809031,"msg":"deRegistered targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2"}
{"level":"info","ts":1665502796.8090775,"msg":"registering targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2","targets":[{"AvailabilityZone":null,"Id":"192.168.16.64","Port":80}]}
{"level":"info","ts":1665502796.952015,"msg":"registered targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2"}
{"level":"info","ts":1665502797.1036577,"msg":"deRegistering targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2","targets":[{"AvailabilityZone":"eu-west-1b","Id":"192.168.83.233","Port":80}]}
{"level":"info","ts":1665502797.151773,"msg":"deRegistered targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2"}
{"level":"info","ts":1665502797.151815,"msg":"registering targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2","targets":[{"AvailabilityZone":null,"Id":"192.168.41.36","Port":80},{"AvailabilityZone":null,"Id":"192.168.64.124","Port":80}]}
{"level":"info","ts":1665502797.338572,"msg":"registered targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2"}
{"level":"info","ts":1665502797.4691834,"msg":"registering targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2","targets":[{"AvailabilityZone":null,"Id":"192.168.48.198","Port":80}]}
{"level":"info","ts":1665502797.6005466,"msg":"registered targets","arn":"arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2"}

同時，過濾 CloudTrail DeregisterTargets 同樣能觀察到使用 AWS Load Balancer Controller IRSA IAM role 更新 Target Group IP 地址資訊。

...
...
    "userIdentity": {
        "type": "AssumedRole",
        "principalId": "AROAYFMQSNSERYVRZKOIQ:1665499421108782293",
        "arn": "arn:aws:sts::111111111111:assumed-role/eksctl-ironman-2022-addon-iamserviceaccount-kube-Role1-MW90E4BAEQWQ/1665499421108782293",
    ...
    ...
    ...
    "eventTime": "2022-10-11T15:39:51Z",
    "eventSource": "elasticloadbalancing.amazonaws.com",
    "eventName": "DeregisterTargets",
    ...
    "userAgent": "elbv2.k8s.aws/v2.4.3 aws-sdk-go/1.42.27 (go1.17.10; linux; amd64)",
    "requestParameters": {
        "targetGroupArn": "arn:aws:elasticloadbalancing:eu-west-1:111111111111:targetgroup/k8s-demongin-serviced-610dd2305f/ffd6c2b5292a36c2",
        "targets": [
            {
                "id": "192.168.22.141",
                "port": 80,
                "availabilityZone": "eu-west-1a"
            },
            {
                "id": "192.168.88.25",
                "port": 80,
                "availabilityZone": "eu-west-1b"
            }
        ]
    }

Kubernetes Service 與 Pod 流程

根據 2019 Kubecon talk [2] Ready? A Deep Dive into Pod Readiness Gates for Service Health... - Minhan Xia & Ping Zou [3]。提及若 Pod 為 Kubernetes Service Endpoint backend 流程：

https://ithelp.ithome.com.tw/upload/images/20221012/20151040KTnzgv51cM.jpg

  1. Kubernetes Control Plane 接收到 client 或是 scaling 機制需要建立 Pod。
  2. Pod 由 kube-scheduler 調度指派至選定 Node 上。
  3. 在指派的 Node 上對應 kubelet 接收 API server 更新（watch verb），並透過 Container Runtime 如 Docker 建立 container。
    1. kubelet 會即時更新 Pod status Ready 回 kube-apiserver
  4. Endpoint Controller 透過 watch verb 接收更新資訊，當 Pod 進入至 Ready 狀態後，則會增加至 Service Object 上 Endpoints list，資訊包含 Pod IP/Port。
  5. kube-proxy 透過 watch verb 接收 API server 更新對應新增 Service IP/Port 資訊至 iptable 路由規則。

為什麼會有 HTTP 5xx 錯誤

https://ithelp.ithome.com.tw/upload/images/20221012/20151040U3GqdqEoul.jpg

來源： Ready? A Deep Dive into Pod Readiness Gates for Service Health... - Minhan Xia & Ping Zou

在第一個 Pod 更新狀態至 Ready 之前，並不會部署第二個 Pod。值得注意的是，我們在更新 Deployment rolling upgrade 過程中，包含了上述步驟 4 集步驟 5 是同步進行。

換言之，在 Deployment Controller 更新下一個 Pod 時，新的 Pod 狀態操作仍在進行中。由於此過程仍會終止舊版本的 Pod，導致新 Pod 已經更新至 Ready 狀態但這些更新仍在進行中並且舊版本的 pod 已終止的情況。

在此次 EKS 環境上使用 AWS Load Balancer Controller 預設並不考慮在 ALB registration times 或是 health checks 時間。

根據上述流程，ALB 已經成功註冊 Pod IP 至 Target group 後，開始將請求轉發給 Pod，然而 Pod 內應用服務仍處於初始化階段，儘管 Pod 狀態更新為 Ready 狀態，但實際仍無法處理請求，而導致 HTTP 500-level error。

反之，當 Pod 被終止時，Pod 仍需等候 Service/Endpoint/Ingress Controller 更新並發送對應 AWS API 進行 registering/deregistering。預設 Kubernetes 也不會設定 ALB Target Group Deregistration delay [4] 屬性，這可能會導致 ALB 仍向已終止的 Pod 的 IP/Port 發送請求。

參考文件

  1. Troubleshoot your Application Load Balancers - HTTP 502: Bad gateway - https://docs.aws.amazon.com/elasticloadbalancing/latest/application/load-balancer-troubleshooting.html#http-502-issues
  2. Ready? A Deep Dive into Pod Readiness Gates for Service Health Management - Minhan Xia, Google & Ping Zou, Intuit | 2019 Kubecon talk - https://kccnceu19.sched.com/event/MPaQ
  3. Ready? A Deep Dive into Pod Readiness Gates for Service Health... - Minhan Xia & Ping Zou - https://www.youtube.com/watch?v=Vw9GmSeomFg
  4. Target groups for your Application Load Balancers - Deregistration delay - https://docs.aws.amazon.com/elasticloadbalancing/latest/application/load-balancer-target-groups.html#deregistration-delay

上一篇
[26] 為什麼 EKS 在更新 Kubernetes deployment 時會有 HTTP 502 的 error（一）
下一篇
[28] 為什麼 EKS 在更新 Kubernetes deployment 時會有 HTTP 502 的 error（三）
系列文
那些文件沒告訴你的 AWS EKS 共 30 篇
  1. 26
    [26] 為什麼 EKS 在更新 Kubernetes deployment 時會有 HTTP 502 的 error（一）
  2. 27
    [27] 為什麼 EKS 在更新 Kubernetes deployment 時會有 HTTP 502 的 error（二）
  3. 28
    [28] 為什麼 EKS 在更新 Kubernetes deployment 時會有 HTTP 502 的 error（三）
  4. 29
    [29] 為什麼 EKS 在更新 Kubernetes deployment 時會有 HTTP 502 的 error（四）
  5. 30
    [30] Summary
完整目錄
圖片
  直播研討會
圖片
{{ item.channelVendor }} {{ item.webinarstarted }} |
{{ formatDate(item.duration) }}
直播中

尚未有邦友留言

立即登入留言
iThome鐵人賽
參賽組數
1064 組
團體組數
40 組
累計文章數
22195 篇
完賽人數
600 人
iT+看影片追技術 看影片追技術 看更多
熱門tag 看更多
15th鐵人賽 16th鐵人賽 13th鐵人賽 14th鐵人賽 12th鐵人賽 11th鐵人賽 鐵人賽 2019鐵人賽 javascript 2018鐵人賽 python 2017鐵人賽 windows php c# windows server linux css react vue.js
熱門問題
熱門回答
熱門文章
IT邦幫忙