Imported from jeremylongshore/tons-of-skills-marketplace (plugins/saas-packs/coreweave-pack/skills/coreweave-data-handling/SKILL.md). Install upstream with npx skills add jeremylongshore/tons-of-skills-marketplace --skill coreweave-data-handling. Copyright stays with the author (MIT).

CoreWeave Data Handling

Community-contributed. Not affiliated with, endorsed by, or sponsored by CoreWeave, Inc. CoreWeave is a registered trademark of CoreWeave, Inc.

Overview

CoreWeave GPU cloud workloads involve large-scale data artifacts: model weights (multi-GB safetensors/GGUF), training datasets (parquet, TFRecord, WebDataset), checkpoint snapshots, and inference cache volumes. Data flows through Kubernetes PersistentVolumeClaims backed by region-specific storage classes. Compliance requires encryption at rest via the storage driver, namespace-scoped RBAC for volume access, and audit logging for any data egress from GPU nodes.

Prerequisites

• Approved data classification, retention schedule, and region for the artifact.
• A namespace-scoped service account, encrypted storage class, and approved destination.
• Expected artifact size and SHA-256 from a trusted source before import.

Instructions

1. Create or select an encrypted PVC in the approved region and grant its mount only to the intended namespace service account.
2. Import artifacts using a short-lived job; verify the expected SHA-256 before any training or serving workload consumes them.
3. Export only to a reviewed destination, preserve the checksum and data-owner approval, and enforce the retention policy for checkpoints and datasets.
4. Record storage provisioning, access changes, deletion, and external egress in the audit system without copying sensitive artifact contents into logs.

Data Classification

Data Import

import { KubeConfig, BatchV1Api } from '@kubernetes/client-node';

async function importDataset(pvcName: string, sourceUrl: string, namespace: string) {
  const kc = new KubeConfig();
  kc.loadFromDefault();
  const batch = kc.makeApiClient(BatchV1Api);
  const job = {
    metadata: { name: `import-${Date.now()}`, namespace },
    spec: { template: { spec: {
      restartPolicy: 'Never',
      containers: [{ name: 'loader', image: 'python:3.11-slim',
        command: ['python3', '-c', `
import urllib.request, hashlib
dest = '/data/dataset.tar.gz'
urllib.request.urlretrieve('${sourceUrl}', dest)
print(f"SHA256: {hashlib.sha256(open(dest,'rb').read()).hexdigest()}")`],
        volumeMounts: [{ name: 'storage', mountPath: '/data' }],
      }],
      volumes: [{ name: 'storage', persistentVolumeClaim: { claimName: pvcName } }],
    }}}
  };
  await batch.createNamespacedJob(namespace, { body: job });
}

Data Export

async function exportCheckpoint(pvcName: string, destBucket: string, ns: string) {
  // Validate export destination is in approved region list
  const APPROVED_REGIONS = ['us-east-1', 'us-central-1', 'eu-west-1'];
  const region = destBucket.split('-').slice(0, 3).join('-');
  if (!APPROVED_REGIONS.some(r => destBucket.includes(r))) {
    throw new Error(`Export blocked: ${region} not in approved regions`);
  }
  // Stream from PVC → object storage with integrity check
  const exportCmd = `tar czf - /models | gsutil cp - gs://${destBucket}/export.tar.gz`;
  console.log(`Exporting from PVC ${pvcName} to ${destBucket}`);
  return exportCmd;
}

Data Validation

interface ModelArtifact {
  name: string; format: 'safetensors' | 'gguf' | 'bin' | 'pt';
  sizeBytes: number; sha256: string;
}

function validateArtifact(artifact: ModelArtifact): string[] {
  const errors: string[] = [];
  if (!artifact.name || artifact.name.length > 255) errors.push('Invalid artifact name');
  if (artifact.sizeBytes <= 0) errors.push('Size must be positive');
  if (!/^[a-f0-9]{64}$/.test(artifact.sha256)) errors.push('Invalid SHA-256 hash');
  if (!['safetensors', 'gguf', 'bin', 'pt'].includes(artifact.format)) errors.push(`Unsupported format`);
  return errors;
}

Compliance

• All PVCs use encrypted storage classes (AES-256 at rest)
• HuggingFace and API tokens stored in Kubernetes Secrets with KMS encryption
• Namespace-scoped RBAC restricts volume mount access to authorized workloads
• Data egress from GPU nodes logged via network policy audit
• Training datasets with PII processed only in approved regions (data residency)
• Checkpoint retention enforced via CronJob garbage collection (30-day default)
• SOC 2 Type II audit trail for all storage provisioning and deletion events

Error Handling

Output

• An encrypted, namespace-scoped storage path with validated artifact integrity.
• A redacted import/export receipt containing source/destination approval, checksum, retention, and data-owner information.
• A reversible failure path that prevents corrupted or unauthorized data from being mounted.

Examples

Import only a manifest-approved artifact and verify its checksum inside the isolated job before promoting it to a serving or training workload:

kubectl -n research apply -f dataset-import-job.yaml
kubectl -n research wait --for=condition=complete job/dataset-import --timeout=30m
kubectl -n research logs job/dataset-import | grep SHA256

If the checksum differs, quarantine the PVC content, retain the redacted job receipt, and reacquire the artifact from the approved source. Do not retry into a production volume or disable integrity verification.

Resources

• CoreWeave Storage Docs
• Kubernetes PVC Reference

Next Steps

See coreweave-security-basics.