volcano-node-resources

Files (2)

• SKILL.md
• get-node-resources.sh · scripts/

SKILL.md (first lines)

# Volcano Node Resources

Query cluster node resources to understand capacity and availability for Volcano scheduling. This skill helps identify resource bottlenecks at the node level.

**Scope:** This skill is for **diagnosis only**. It retrieves resource information but does not modify any cluster state.

## Usage

```bash
bash skills/core/volcano-node-resources/scripts/get-node-resources.sh [options]
```

## Parameters

| Parameter | Required | Description |
|-----------|----------|-------------|
| `--node NODE` | no | Query specific node only |
| `--label LABEL` | no | Filter nodes by label (e.g., gpu=true) |
| `--show-usage` | no | Show current resource usage (requires metrics-server) |
| `--show-pods` | no | Show pods running on each node |
| `--format FORMAT` | no | Output format: table (default), json, wide |

## Examples

Get overview of all nodes:
```bash
bash skills/core/volcano-node-resources/scripts/get-node-resources.sh
```

Check specific node:
```bash
bash skills/core/volcano-node-resources/scripts/get-node-resources.sh --node worker-1
```

Check GPU nodes:
```bash
bash skills/core/volcano-node-resources/scripts/get-node-resources.sh --label nvidia.com/gpu.present=true
```

Show resource usage:
```bash
bash skills/core/volcano-node-resources/scripts/get-node-resources.sh --show-usage
```

Show with pod information:
```bash
bash skills/core/volcano-node-resources/scripts/get-node-resources.sh --show-pods
```

JSON output for parsing:
```bash
bash skills/core/volcano-node-resources/scripts/get-node-resources.sh --format json
```

## Understanding Node Resources

### Resource Types

| Resource | Description | Scheduling Impact |
|----------|-------------|-------------------|
| `cpu` | CPU cores (millicores) | Primary scheduling constraint |
| `memory` | RAM (bytes) | Primary scheduling constraint |
| `nvidia.com/gpu` | GPU devices | Hardware-specific scheduling |
| `pods` | Max pods per node | Density limit |
| `ephemeral-storage` | Disk space | Secondary constraint |

### Capacity vs Allocatable

- **Capacity**: Total physical resources on the node
- **Allocatable**: Resources available for pods (capacity minus system reservations)

**Reservations include:**
- kubelet overhead
- System daemons (kube-proxy, node-exporter)
- Kernel reserved memory
- Eviction threshold

### Resource Usage

When `--show-usage` is enabled and metrics-server is available:

- **Requests**: Sum of all pod resource requests on the node
- **Limits**: Sum of all pod resource limits
- **Usage**: Actual resource consumption

**Key insights:**
- Allocatable - Requests = Available for new pods
- Usage < Requests = Over-provisioning
- Usage > Requests = Over-committing (risky)

## Resource Calculation

### Available Resources

```
Available = Allocatable - Allocated (sum of all requests)
```

Nodes with zero or negative available resources cannot accept new pods.

### Gang Scheduling Calculation

For Gang scheduling, you need:
```
Number of nodes with Available >= Pod Request >= minMember
```

Example:
- minMember = 4
- Pod requests 4 CPUs each
- Need at least 4 nodes with 4+ CPUs available

## Diagnostic Use Cases

### Case 1: Identify Nodes with Available Resources

```bash
bash skills/core/volcano-node-resources/scripts/get-node-resources.sh
```

From scitix/siclaw, Apache-2.0.

More from scitix/siclaw