How to configure container startup and exit order-Elastic Container Instance(ECI)-阿里云帮助中心

How it works

By default, containers in an ECI pod start and exit in parallel, without a specific order. Some scenarios involve container dependencies, requiring one container to start only after another is running, or to exit only after another has stopped. For example:

In an Istio service governance scenario, the istio-proxy container must be Ready before the application container generates traffic, and it must exit only after the application container exits.
For log collection, the log agent container must be Ready before the application container produces logs, and it must exit only after the application container exits.

To address these scenarios, ECI allows you to set startup and exit priorities for containers by using environment variables, ensuring they start and exit in a specified order.

Configuration

The following environment variables control the container startup and exit order:

Parameter	Environment variable	Description
Container startup priority	ECI_CONTAINER_LAUNCH_PRIORITY	Valid values: -1000 to 1000. Default: 0. A larger value indicates a higher startup priority. A container with a higher priority always starts before a container with a lower priority. Containers with the same priority start in parallel, and their startup order is not guaranteed. If a container (for example, B) depends on a service from another container (A), you must set a higher startup priority for container A and also configure a readiness probe for it. This probe ensures that container B starts only after container A is ready.
Container exit priority	ECI_CONTAINER_EXIT_PRIORITY	Valid values: -1000 to 1000. Default: 0. A larger value indicates a higher exit priority. A container with a higher priority always exits before a container with a lower priority. Containers with the same priority exit in parallel, and their exit order is not guaranteed.

Important

Configuring the container exit order may cause the pod's total termination time to exceed the TerminationGracePeriodSeconds value in its pod specification.

Examples

Configure container startup order

Use the following command to create a Deployment from a YAML configuration file.

kubectl apply -f test-launch.yaml

The following test-launch.yaml content defines a Deployment with one replica. The pod contains two containers, c1 and c2, where c1 has a higher startup priority than c2. A readiness probe is configured for c1 to ensure that c2 starts only after c1 is ready.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: test-launch
  labels:
    app: test
spec:
  replicas: 1 
  selector:
    matchLabels:
      app: test
  template:
    metadata:
      labels:
        app: test
    spec:
      containers:
      - image: registry-vpc.cn-shanghai.aliyuncs.com/eci_open/nginx:alpine
        name: c1
        env:
        - name: ECI_CONTAINER_LAUNCH_PRIORITY
          value: "1000"
        readinessProbe:                  
          httpGet:
            path: /
            port: 80
          initialDelaySeconds: 30     
          periodSeconds: 3
      - image: registry-vpc.cn-shanghai.aliyuncs.com/eci_open/nginx:alpine
        name: c2
        env:
        - name: ECI_CONTAINER_LAUNCH_PRIORITY
          value: "0"
        args:
        - /bin/sh
        - -c
        - sleep 3600s

Check the container startup times in the pod's status.

kubectl describe pod <pod name>

The Containers section of the output shows that the c2 container starts after the c1 container. An example follows:

Containers:
c1:
    Container ID:   containerd://779ec5b6d9ad1164c929ab0b8bbc4a3a3b24fe1f9654867e842f2dcfe7b24beb
    Image:          registry.cn-shanghai.aliyuncs.com/eci_open/nginx:alpine
    Image ID:       registry.cn-shanghai.aliyuncs.com/eci_open/nginx@sha256:2ec3c026183996087f26c6b3
    Port:           <none>
    Host Port:      <none>
    State:          Running
      Started:      Wed, 16 Aug 2023 07:11:48 +0000
    Ready:          True
    Restart Count:  0
    Readiness:      http-get http://:80/ delay=30s timeout=1s period=3s #success=1 #failure=3
    Environment:
      ECI_CONTAINER_LAUNCH_PRIORITY:  1000
    Mounts:
      /var/run/secrets/kubernetes.io/serviceaccount from kube-api-access-2qkqp (ro)
c2:
    Container ID:   containerd://fe6261cc2e659ee9afb07c2efd516250aeb854544504d66ac011b557e94a0cb6
    Image:          registry-vpc.cn-shanghai.aliyuncs.com/eci_open/nginx:alpine
    Image ID:       registry.cn-shanghai.aliyuncs.com/eci_open/nginx@sha256:2ec3c026183996087f26c6b38
    Port:           <none>
    Host Port:      <none>
    Args:
      /bin/sh
      -c
      sleep 3600s
    State:          Running
      Started:      Wed, 16 Aug 2023 07:12:20 +0000
    Ready:          True
    Restart Count:  0
    Environment:
      ECI_CONTAINER_LAUNCH_PRIORITY:  0

Configure container exit order

Use the following command to create a Deployment from a YAML configuration file.

kubectl apply -f test-exit.yaml

The following test-exit.yaml content defines a Deployment with one replica. The pod contains three containers, with exit priorities set in descending order: c1, c2, and c3.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: test-exit
  labels:
    app: test
spec:
  replicas: 1 
  selector:
    matchLabels:
      app: test
  template:
    metadata:
      labels:
        app: test
    spec:
      containers:
      - image: registry-vpc.cn-shanghai.aliyuncs.com/eci_open/nginx:alpine
        name: c1
        env:
        - name: ECI_CONTAINER_EXIT_PRIORITY
          value: "1000"
      - image: registry-vpc.cn-shanghai.aliyuncs.com/eci_open/nginx:alpine
        name: c2
        env:
        - name: ECI_CONTAINER_EXIT_PRIORITY
          value: "0"
        args:
        - /bin/sh
        - -c
        - sleep 3600s
      - image: registry-vpc.cn-shanghai.aliyuncs.com/eci_open/nginx:alpine
        name: c3
        env:
        - name: ECI_CONTAINER_EXIT_PRIORITY
          value: "-1000"
        args:
        - /bin/sh
        - -c
        - sleep 3600s

Delete the pod and then observe the order of the Killing events for the containers.

Note

These events are transient and may be unavailable after the pod is deleted.

kubectl describe pod <pod name>

The Events section of the output shows the containers stopping in order of their exit priority: c1, c2, and then c3. An example follows:

Events:
  Type     Reason                  Age    From               Message
  ----     ------                  ----   ----               -------
  Warning  MissingClusterDNS       2m10s  virtual-kubelet    pod: default/test-exit-pr
and cannot create Pod using "ClusterFirst" policy. Falling back to "Default" policy.
  Normal   DefaultInstanceTypeMatch  2m10s  EciService       [eci.containergroup]The de
  Warning  ImageCacheMissed        2m9s   EciService         [eci.imagecache]Missed ima
  Normal   ImageCacheAutoCreated   2m9s   EciService         [eci.imagecache]Image cach
  Normal   Pulling                 115s   kubelet            Pulling image "registry.cn
  Normal   Pulled                  113s   kubelet            Successfully pulled image
ing)
  Normal   Created                 113s   kubelet            Created container c1
  Normal   Started                 113s   kubelet            Started container c1
  Normal   Pulled                  113s   kubelet            Container image "registry.
  Normal   Created                 113s   kubelet            Created container c2
  Normal   Started                 113s   kubelet            Started container c2
  Normal   Pulled                  113s   kubelet            Container image "registry.
  Normal   Created                 112s   kubelet            Created container c3
  Normal   Started                 112s   kubelet            Started container c3
  Normal   Killing                 37s    kubelet            Stopping container c1
  Normal   Killing                 37s    kubelet            Stopping container c2
  Normal   Killing                 7s     kubelet            Stopping container c3