### Transformation constraints

In order for the framework to properly handle dependencies and events, transformation functions must adhere by a few properties.

Basically, Transformations must be stateless and idempotent.
* Any querying of other `Collection`s _must_ be done through `krt.Fetch`.
* Querying other data stores that may change is not permitted.
* Querying external state (e.g. making HTTP calls) is not permitted.

 * Immutable empty instances are returned for collection types; {@code ""} for string; {@code 0} for
 * number types; reasonable default instance for other stateless types. For mutable types, a fresh
 * instance is created each time {@code get()} is called.
 *
 * @author Kevin Bourrillion
 * @author Ben Yu
 * @since 12.0
 */
@GwtIncompatible
@J2ktIncompatible

use the gob encoding, an efficient, robust, self-describing binary
protocol. A fact type may implement the GobEncoder/GobDecoder interfaces
if the default encoding is unsuitable. Facts should be stateless.
Because serialized facts may appear within build outputs, the gob encoding
of a fact must be deterministic, to avoid spurious cache misses in
build systems that use content-addressable caches.

//
// The op built by this builder is stateful if its behavior depends on some
// state beyond its input tensors (e.g. variable reading op) or if it has a
// side-effect (e.g. printing or asserting ops). Equivalently, stateless ops
// must always produce the same output for the same input and have no
// side-effects.
//
// By default Ops may be moved between devices. Stateful ops should either not

 * Immutable empty instances are returned for collection types; {@code ""} for string; {@code 0} for
 * number types; reasonable default instance for other stateless types. For mutable types, a fresh
 * instance is created each time {@code get()} is called.
 *
 * @author Kevin Bourrillion
 * @author Ben Yu
 * @since 12.0
 */
@GwtIncompatible
@J2ktIncompatible

## **1. Cloud-native Architecture**

![cloud-native](https://github.com/minio/minio/blob/master/docs/bigdata/images/image1.png?raw=true "cloud native architecture")

Kubernetes manages stateless Spark and Hive containers elastically on the compute nodes. Spark has native scheduler integration with Kubernetes. Hive, for legacy reasons, uses YARN scheduler on top of Kubernetes.

    }
  }

  // In theory we can use PartitionedCall if the XLA cluster does not have any
  // stateful operations.  However, for now we choose to be conservative since
  // we don't have any evidence that choosing a stateless partitioned call helps
  // for performance.
  ops::StatefulPartitionedCall call(
      root.WithOpName("stateful_partitioned_call"), args, n->output_types(),

	// Cgroup subsystem to their mount location.
	// e.g.: "cpu" -> "/sys/fs/cgroup/cpu"
	MountPoints map[string]string
}

// cgroupManagerImpl implements the CgroupManager interface.
// Its a stateless object which can be used to
// update,create or delete any number of cgroups
// It relies on runc/libcontainer cgroup managers.
type cgroupManagerImpl struct {
	// subsystems holds information about all the

	Supported language.Coverage

	// We keep the following lists separate, instead of having a single per-
	// language struct, to give the compiler a chance to remove unused code.

	// Some uppercase mappers are stateless, so we can precompute the
	// Transformers and save a bit on runtime allocations.
	upperFunc = []struct {
		upper mapFunc
		span  spanFunc
	}{
		{nil, nil},                  // und

	if err != nil {
		return newClaim, err
	}
	_, err = ctrl.storeClaimUpdate(logger, newClaim)
	if err != nil {
		return newClaim, err
	}
	return newClaim, nil
}

// Stateless functions

func getClaimStatusForLogging(claim *v1.PersistentVolumeClaim) string {
	bound := metav1.HasAnnotation(claim.ObjectMeta, storagehelpers.AnnBindCompleted)