Resources

Learn how to assign CPU/memory resources to your containerized application.
This section continues from the previous section - make sure you do the tutorial in sequence.
Deployment
/deployment/kubernetes/deployment
Default Configuration
You can specify the computing resource needs for each of the containers. By default, each container is given 10% of a CPU and no memory use restrictions.
The defaults can cause issues:
If a Node has 1 full CPU, then Kubernetes may schedule up to 10 instances of the same container, which may overload the system.
If a Node has 16GB of RAM, and without memory restriction, then each container instance (JVM) may think they each can use up to 16GB, causing memory overuse (and thus, virtual memory swapping, etc)
You can see the current resource by describing a Pod instance, look for the Requests/Limits lines.
POD_NAME=$(kubectl get pods -lapp=helloworld -o jsonpath='{.items[0].metadata.name}')
​
kubectl describe pod $POD_NAME
The details should have a Requests section with cpu value set to 100m:
Name:           helloworld-...
Namespace:      default...
Containers:
  helloworld:
    ...
    Requests:
      cpu:  100m
...
The default value is 100m, which means 100 milli = 100/1000 = 10%of a vCPU core.
The default is configured per Namespace. The application was deployed into the default Namespace. Look at the default resource configuration for this Namespace:
kubectl describe ns default
See the output:
Name:         default
Labels:       <none>
Annotations:  <none>
Status:       Active
​
Resource Quotas
 Name:                       gke-resource-quotas
 Resource                    Used  Hard
 --------                    ---   ---
 count/ingresses.extensions  1     100
 count/jobs.batch            0     5k
 pods                        3     1500
 services                    2     500
​
Resource Limits
 Type       Resource  Min  Max  Default Request  Default Limit  Max Limit/Request Ratio
 ----       --------  ---  ---  ---------------  -------------  -----------------------
 Container  cpu       -    -    100m             -              -
However, the configuration is actually stored in a LimitRange Kubernetes resource:
kubectl get limitrange limits -oyaml
The default can be updated. See Configure Default CPU Requests and Limits for a Namespace documentation.
Resource Request
In Kubernetes, you can reserve capacity by setting the Resource Requests to reserve more CPU and memory. Configure the deployment to reserve at least 20% of a CPU, and 128Mi of RAM.
k8s/deployment.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  labels:
    app: helloworld
  name: helloworld
spec:
  replicas: 1
  selector:
    matchLabels:
      app: helloworld
  template:
    metadata:
      labels:
        app: helloworld
    spec:
      containers:
      - image: gcr.io/.../helloworld
        name: helloworld
        # Add the resources requests block
        resources:
          requests:
            cpu: 200m
            memory: 128Mi
In this example, CPU request is 200m which means 200 milli=200/1000 = 20% of 1 vCPU core.
Memory is 128Mi, which is 128 Mebibytes = ~134 Megabytes.
See Kubernetes Resource Units documentation for the units descriptions such as m, M, and Mi.
When specifying the Memory resource allocation, do not accidentally use m as the unit. 128m means 0.128 bytes.
Resource Limit
The application can consume more CPU and memory than requested - it can burst up to the limit, but cannot exceed the limit. Configure the deployment to set the limit:
k8s/service.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  labels:
    app: helloworld
  name: helloworld
spec:
  replicas: 1
  selector:
    matchLabels:
      app: helloworld
  template:
    metadata:
      labels:
        app: helloworld
    spec:
      containers:
      - image: gcr.io/.../helloworld
        name: helloworld
        # Add the resources requests block
        resources:
          requests:
            cpu: 200m
            memory: 256Mi
          limits:
            cpu: 500m
            memory: 256Mi
CPU limit is a compressible resource. If the application exceeds the CPU limit, it'll simply be throttled, and thus capping the latency and throughput.
Memory is not a compressible resource. If the application exceeds the Memory limit, then the container will be killed (OOMKilled) and restarted.
For Java applications, read the Container Awareness section to make sure you are using a Container-Aware OpenJDK version to avoid unnecessary OOMKilled errors.
Deployment
Service
Last updated 5 months ago