Container Service for Kubernetes:Best practices for ACK managed clusters and ACK dedicated clusters

Edition

Reference

LLM inference

• Quickly deploy a model in ACK

• Use vLLM to deploy a Qwen model as an inference service

• Use rtp-llm to deploy a Qwen model as an inference service

• Use Triton to deploy a Qwen model as an inference service

• Use TGI to deploy a Qwen model as an inference service

AIGC

• Build AI images and accelerate image pulling

• Deploy a Stable Diffusion Service based on Knative

• Use CPUs to accelerate text-to-image inference for SDXL Turbo

Authorization

Best practices for authorization

Clusters

• Suggestions on choosing ECS specifications for ACK clusters

• Suggestions on building an HA cluster

• Plan CIDR blocks for an ACK cluster

• Suggestions on using large ACK Pro clusters

• Increase the capacity of etcd for dedicated Kubernetes clusters

Nodes and node pools

• Best practices for associating deployment sets with node pools

• Best practices for preemptible instance-based node pools

• Use ContainerOS to quickly scale out nodes

• Add free nodes to a node pool

• Comparison of Docker, containerd, and Sandboxed-Container

• Create cgroup v2 nodes from a custom image

Networks

• Improve the performance of the NetworkPolicy feature for a large ACK cluster in Terway mode

• Use the Gateway API to access Services

• Best practices for DNS services

• Usage notes of the NGINX Ingress controller

Application management

• Recommended configurations for workloads

• Build AI images and accelerate image pulling

• Set up Jenkins to build an application delivery pipeline

• Use GitLab CI to run a GitLab runner and run a pipeline on Kubernetes

• Elastic and cost-effective CI/CD based on ACK Serverless

Knative

• View alert events of Services

• Enable Log Service on Knative

• Use elastic container instances in Knative

• Use HPA in Knative

• Enable automatic scaling for pods based on the number of requests

• View the Knative dashboard in Prometheus Service

• Deploy a Stable Diffusion Service based on Knative

Storage

• Use a Deployment to create ephemeral volumes

• Recommended storage settings for cross-zone deployment

• Best practice for throttling block storage devices

• Best practices for data security of disk volumes

• Migrate stateful applications that use disk volumes across zones

• Use the multi-attach and NVMe reservation features of NVMe disks

• Enable a StatefulSet to support persistent storage

• Best practice for OSS read/write splitting

• Use CNFS to automatically collect the heap dumps of a JVM

Observability

• Use ACK Net Exporter to troubleshoot network issues

• Use a Prometheus client to monitor applications

• Best practices for configuring alert rules in Prometheus

• Use SysOM to locate container memory issues

Cost management suite

Use the cost insights feature to manage and optimize costs

Auto scaling

• Use ack-autoscaling-placeholder to scale pods within seconds

• Create custom images

• Configure auto scaling for cross-zone deployment

• Use the UnitedDeployment controller in ACK clusters

• Use ARMS APM to enable horizontal pod autoscaling

• Enable auto scaling based on GPU metrics

Service Mesh (ASM)

• Use ASM to manage applications in Elastic Container Instance-based pods running on virtual nodes of ACK clusters

• Manage applications in a registered Kubernetes cluster by using ASM

• Use an ASM serverless gateway to act as a single entry point for access to multiple clusters

• Implement cross-region disaster recovery and load balancing by using multiple clusters

• Implement auto scaling for workloads by using ASM metrics

• Use an ingress gateway to access a gRPC service in an ASM instance

Graphics applications

• Best practice for Linux graphics applications

• Best practice for Unity remote rendering