tenant isolation, resource allocation, model sharing and hardware attacks.

### Tenant isolation

Since any tenants or users providing models, graphs or checkpoints can execute
code in context of the TensorFlow service, it is important to design isolation
mechanisms that prevent unwanted access to the data from other tenants.

Install MinIO on Kubernetes:

- [MinIO Quickstart Guide for Kubernetes](https://min.io/docs/minio/kubernetes/upstream/index.html#quickstart-for-kubernetes).
- [Deploy a Tenant from the MinIO Operator](https://min.io/docs/minio/kubernetes/upstream/operations/install-deploy-manage/deploy-minio-tenant.html)

Install Distributed MinIO on Linux:

# MinIO Server Limits Per Tenant

For optimal production setup MinIO recommends Linux kernel version 4.x and later.

## Erasure Code (Multiple Drives / Servers)

| Item                                                            | Specification |
|:----------------------------------------------------------------|:--------------|
| Maximum number of servers per cluster                           | no-limit      |

| This Helm chart is community built, maintained, and supported. MinIO does not guarantee support for any given bug, feature request, or update referencing this chart. <br/><br/> MinIO publishes a separate [MinIO Kubernetes Operator and Tenant Helm Chart](https://github.com/minio/operator/tree/master/helm) that is officially maintained and supported. MinIO strongly recommends using the MinIO Kubernetes Operator for production deployments. See [Deploy Operator With Helm](https://min.io/...

[here](https://github.com/minio/minio/blob/master/docs/kms/IAM.md).

> NOTE: if *path_prefix* is set then MinIO will not federate your buckets, namespaced IAM assets are assumed as isolated tenants, only buckets are considered globally unique but performing a lookup with a *bucket* which belongs to a different tenant will fail unlike federated setups where MinIO would port-forward and route the request to relevant cluster accordingly. This is a special feature, federated deployments should not need to...

	sys.hMutex.RLock()
	defer sys.hMutex.RUnlock()
	m := make(map[string]epHealth, len(sys.hc))
	for k, v := range sys.hc {
		m[k] = v
	}
	return m
}

// ListTargets lists bucket targets across tenant or for individual bucket, and returns
// results filtered by arnType
func (sys *BucketTargetSys) ListTargets(ctx context.Context, bucket, arnType string) (targets []madmin.BucketTarget) {
	h := sys.healthStats()

						globalDomainNames, err))
				}
			}
		}
	}

	// Bucket federation is 'true' only when IAM assets are not namespaced
	// per tenant and all tenants interested in globally available users
	// if namespace was requested such as specifying etcdPathPrefix then
	// we assume that users are interested in global bucket support
	// but not federation.

    protected static final String AZUREAD_STATE_TTL = "aad.state.ttl";

    protected static final String AZUREAD_AUTHORITY = "aad.authority";

    protected static final String AZUREAD_TENANT = "aad.tenant";

    protected static final String AZUREAD_CLIENT_SECRET = "aad.client.secret";

    protected static final String AZUREAD_CLIENT_ID = "aad.client.id";

	}
	return sys.collectPeerMetrics(ctx, peerChannels, g)
}

// ServiceFreeze freezes all S3 API calls when 'freeze' is true,
// 'freeze' is 'false' would resume all S3 API calls again.
// NOTE: once a tenant is frozen either two things needs to
// happen before resuming normal operations.
//   - Server needs to be restarted 'mc admin service restart'
//   - 'freeze' should be set to 'false' for this call

			return
		}
	}

	for _, bucket := range bucketsInConflict.ToSlice() {
		dnsLogIf(ctx, fmt.Errorf("Unable to add bucket DNS entry for bucket %s, an entry exists for the same bucket by a different tenant. This local bucket will be ignored. Bucket names are globally unique in federated deployments. Use path style requests on following addresses '%v' to access this bucket", bucket, globalDomainIPs.ToSlice()))
	}