The security impact of skipping those checks should be low, since the attack
scenario would require a malicious user to be able to control the model which as
stated above is already equivalent to code execution. In any case, the
recommendation is not to serve models using Eager mode since it also has
performance limitations.

## Multi-Tenant environments

It is possible to run multiple TensorFlow models in parallel. For example,

  // Standard object's metadata.
  // The name has no particular meaning. It must be a DNS subdomain (dots allowed, 253 characters).
  // To ensure that there are no conflicts with other CSI drivers on the cluster,
  // the recommendation is to use csisc-<uuid>, a generated name, or a reverse-domain name
  // which ends with the unique CSI driver name.
  //
  // Objects are namespaced.
  //

  // Standard object's metadata. The name has no particular meaning. It must be
  // be a DNS subdomain (dots allowed, 253 characters). To ensure that
  // there are no conflicts with other CSI drivers on the cluster, the recommendation
  // is to use csisc-<uuid>, a generated name, or a reverse-domain name which ends
  // with the unique CSI driver name.
  //
  // Objects are namespaced.
  //

## Known Issues

### Etcd v3.5.[0-2] data corruption

Data corruption issue was found in etcd v3.5.0 release that was shipped with 1.22 Kubernetes release. Please read up-to-date [production recommendations for etcd](https://github.com/etcd-io/etcd/tree/main/CHANGELOG).

## Changes by Kind

### Deprecation

### Etcd v3.5.[0-2] data corruption

Data corruption issue was found in etcd v3.5.0 release that was shipped with 1.22 Kubernetes release. Please read up-to-date [production recommendations for etcd](https://github.com/etcd-io/etcd/tree/main/CHANGELOG).

## Urgent Upgrade Notes 

### (No, really, you MUST read this before you upgrade)

     * exceptions' docs suggest that either is acceptable. Google's Java Practices page recommends
     * IllegalArgumentException here, in part to keep its recommendation simple: Static methods
     * should throw IllegalStateException only when they use static state.
     *
     * Why do we deviate here? The answer: We want for fluentFuture.getDone() to throw the same

    states. Specifically, it no longer ever drops the `c` (memory) state of an
    `LSTMStateTuple`. The new behavior leads to proper dropout behavior for
    LSTMs and stacked LSTMs. This bug fix follows recommendations from published
    literature, but is a behavioral change. State dropout behavior may be
    customized via the new `dropout_state_filter_visitor` argument.

- Replace scale down forbidden window with scale down stabilization window. Rather than waiting a fixed period of time between scale downs HPA now scales down to the highest recommendation it during the scale down stabilization window. ([#68122](https://github.com/kubernetes/kubernetes/pull/68122), [@krzysztof-jastrzebski](https://github.com/krzysztof-jastrzebski))

## External Dependency Version Information

Continuous integration builds have used the following versions of external dependencies, however, this is not a strong recommendation and users should consult an appropriate installation or upgrade guide before deciding what versions of etcd, docker or rkt to use.

* Docker versions 1.10.3 - 1.12.3
  * Docker version 1.11.2 known issues

## **External Dependency Version Information**

Continuous integration builds have used the following versions of external dependencies, however, this is not a strong recommendation and users should consult an appropriate installation or upgrade guide before deciding what versions of etcd, docker or rkt to use.

* Docker versions 1.10.3, 1.11.2, 1.12.6 have been validated