> ## Documentation Index
> Fetch the complete documentation index at: https://docs.scitix.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Select and Install Cluster Components

> Determine the additional cluster capabilities needed based on workload dependencies and operational goals, then install or maintain the corresponding components.

## Component Overview

Components extend CKS clusters with storage, network, device, scheduling, training, and O\&M capabilities. Do not install components in batches based solely on their names. First, determine which capabilities your workloads lack:

* When workloads require network, persistent storage, GPU, RDMA, or node management capabilities, see the foundational capability components.
* When you need to expose services, run batch or distributed training jobs, or orchestrate workflows, see the business runtime components.
* When you need to collect metrics, logs, and events, or perform inspections, backups, and multi-cluster management, see the observability and operations components.

### Prepare Basic Capabilities for Workloads

This set of components provides the cluster with container networking, persistent storage, node management, and device resources. If a workload depends on any of these capabilities, ensure that the corresponding component is installed first.

#### Cluster Network

The cluster network component provides network communication and network policy capabilities for Pods.

| Component | Function | When to Install |
| - | - | - |
| `cilium` | Provides eBPF CNI container networking and network policy capabilities. | Required when you need to use Cilium networking capabilities or network policies. |
| `calico` | Provides BGP/IPIP CNI container networking and network policy capabilities. | Required when you need to use Calico networking capabilities or network policies. |

#### Storage

The storage component enables workloads to mount platform storage resources.

| Component | Function | When to install |
| - | - | - |
| `scitix-csi-plugin` | Provides CSI drivers for block storage or file storage. | Required when mounting platform storage resources for workloads. |

#### Node Management

The node management component is used to manage node groups and node labels.

| Component | Function | When to install |
| - | - | - |
| `navix-nodegroup-controller` | Provides elastic control capabilities for node groups. | When you need to manage or elastically scale node resources by node group. |
| `scitix-label-manager` | Manages node labels. | When you need to maintain node labels in batches or automatically. |

#### Device Resources

The device resource component registers GPU or RDMA devices as schedulable Kubernetes resources, enabling workloads to request and use the corresponding devices.

| Component | Purpose | When to install |
| - | - | - |
| `nvidia-device-plugin` | Reports and allocates NVIDIA GPU device resources. | Required when requesting and using GPUs in Pods. |
| `rdma-device-plugin` | Reports and allocates shared RDMA NIC device resources. | Required when using RDMA network capabilities in workloads. |

### Supported Service Operating Modes

These components provide traffic entry points for services, run batch or distributed training jobs, and orchestrate and maintain application workflows.

#### Traffic Entry Points

The traffic entry component exposes services in the cluster externally through a Gateway or Ingress.

| Component | Function | When to install |
| - | - | - |
| `envoyproxy-gateway` | Provides an Envoy gateway and a north-south traffic entry point. | When you need to manage ingress traffic through the Gateway API. |
| `nginx-ingress-controller` | Provides an NGINX Ingress traffic entry point. | When you need to expose HTTP or HTTPS services through Ingress. |

#### Scheduling and Training

Scheduling and training components support batch scheduling, group scheduling, and distributed training jobs.

| Component | Function | When to install |
| - | - | - |
| `volcano` | Provides batch and gang scheduling capabilities. | Required for running AI training, batch processing, or multi-Pod coordinated workloads. |
| `training-operator` | Provides an operator for distributed training. | Required for running distributed training workloads such as PyTorchJob. |

#### Workflows and Application Operations and Maintenance

Workflows and application O\&M components are used to orchestrate multi-step tasks or enhance workload O\&M capabilities.

| Component | Purpose | When to install |
| - | - | - |
| `argo-workflow` | Provides a workflow orchestration engine. | When you need to orchestrate multi-step training, data processing, or automated tasks. |
| `kruise` | Provides enhanced workload management capabilities. | When you need enhanced application O\&M capabilities, such as in-place upgrades. |

### Monitor and Maintain Clusters

This set of components is used to collect metrics, logs, and events; detect cluster anomalies; perform checks or backups; and manage multiple clusters.

#### Monitor Metrics

The monitoring metrics component collects metrics from Kubernetes objects, nodes, storage, GPUs, CPUs, network hardware, the control plane, and high-availability components.

| Component | Purpose | When to install |
| - | - | - |
| `metrics-server` | Collects Pod and node resource metrics. | When you need to view basic resource usage or use HPA capabilities that depend on resource metrics. |
| `kube-state-metrics` | Collects Kubernetes object status metrics. | When you need to monitor the status of Kubernetes objects such as Deployments, Pods, and Nodes. |
| `node-exporter` | Collects node system metrics. | When you need to view system metrics for nodes, such as CPU, memory, disks, and network. |
| `gpfs-exporter` | Collects GPFS storage metrics. | When you need to view GPFS file system capacity, performance, or operational status metrics. |
| `dcgm-exporter` | Collects GPU metrics. | When you need to monitor GPU utilization, GPU memory, temperature, and other GPU operating conditions. |
| `pcm-exporter` | Collects CPU PCM performance metrics. | When you need to analyze underlying performance data such as CPU performance or memory bandwidth. |
| `ib-hca-exporter` | Collects InfiniBand network adapter metrics. | When you need to monitor InfiniBand network status and performance. |
| `ipmi-exporter` | Collects out-of-band IPMI hardware metrics. | When you need to monitor server hardware health status. |
| `haproxy-exporter` | Collects control plane HAProxy metrics. | When you need to monitor the status of control plane proxy services. |
| `keepalived-exporter` | Collects VIP keepalived status metrics. | When you need to monitor the high-availability status of control plane or ingress VIPs. |
| `vmo` | Provides victoria-metrics-operator. | When you need to extend monitoring capabilities through the VictoriaMetrics ecosystem. |
| `prometheus-monitoring-cks` | Provides Prometheus capabilities for Kubernetes monitoring. | When you need to deploy Prometheus monitoring for CKS clusters. |

#### Logs and Events

The logs and events component is used to collect cluster logs, Kubernetes events, or telemetry data.

| Component | Purpose | When to Install |
| - | - | - |
| `logging-stack` | Provides a log collection pipeline. | When you need to collect and query cluster or workload logs. |
| `event-exporter` | Collects and reports Kubernetes events. | When you need to track Kubernetes events such as scheduling, startup, failures, and evictions. |
| `otel-collector` | Collects and forwards telemetry data. | When you need to integrate with OpenTelemetry-related observability pipelines. |

#### Cluster Operations and Maintenance

Cluster O\&M components are used to detect node or cluster exceptions and support inspection, self-healing, and backup.

| Component | Purpose | When to install |
| - | - | - |
| `node-problem-detector` | Detects node kernel, GPU, and other failures. | When you need to detect node-level failures and help troubleshoot issues. |
| `aegis-cks` | Provides an Agent for fault self-healing inspection. | When you need to perform inspection and self-healing-related O\&M for CKS clusters. |
| `aegis` | Provides an intelligent O\&M framework. | When you need to use the intelligent O\&M capabilities provided by the platform. |
| `cks-inspect` | Provides automated checks for additional cluster configurations. | When you need to check cluster configurations or assist with validating cluster status. |
| `etcd-backup` | Backs up etcd to object storage on a scheduled basis. | When you need to configure periodic backups for etcd data. |

#### Multi-Cluster Management

The multi-cluster management component is used to centrally manage multiple CKS clusters.

| Component | Purpose | When to install |
| - | - | - |
| `karmada` | Provides a multi-cluster federation control plane. | When you need to centrally manage multiple CKS clusters. |

## Install Components

After identifying the components required for the workload or O\&M objective, proceed with installation. Installing multiple components with similar functions may increase resource consumption and maintenance complexity.

1. Log in to the [GPU Cloud console](https://console.scitix.ai/gpu-cloud/home) and go to the **Clusters** page.

2. Click the target cluster name to access the cluster workspace.

3. In the left navigation pane, go to the **Component** page.

4. Click **Install**.

   ![Install Cluster Components](https://mintcdn.com/scitix-adfc65c2/zAestxkyOFggIEE8/gpu-cloud/cluster/media/cluster-component-install.png?fit=max\&auto=format\&n=zAestxkyOFggIEE8\&q=85\&s=3f033a5e868e6629873fae7fb543f460)

5. In the **Install component** dialog box, select a component.

   When the component list is long, enter a component name in the search box to filter the list.

6. After confirming that the component name and purpose meet business requirements, submit the installation.

7. Return to the **Component** page. After confirming that the component status is normal, deploy workloads that depend on this component.

   If a component is in an abnormal state, first verify that the cluster resources, namespace, storage, or network configuration it depends on are functioning properly. If the cause still cannot be identified, record the component name, version, namespace, and error message, and contact technical support.

## Update or remove components

> [!WARNING]
>
> Updating or deleting a component may affect business operations, monitoring, logs, storage, network, or O\&M processes that depend on it. Before proceeding, confirm the component’s purpose, dependencies, and scope of impact.

1. On the **Clusters** page, click the name of the target cluster.

2. In the left navigation pane, go to the **Component** page.

3. Locate the target component and confirm its purpose, as well as the workloads or operational processes that depend on it.

4. Perform the following actions as needed:

   * Update a component: Click **Update**.
   * Delete a component: Click **Delete** and confirm as prompted.

5. After completing the operation, return to the **Component** page to verify the component status.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.