//     Those cannot be well described when embedded.
//  3. Inconsistent validation.  Because the usages are different, the validation rules are different by usage, which makes it hard for users to predict what will happen.
//  4. The fields are both imprecise and overly precise.  Kind is not a precise mapping to a URL. This can produce ambiguity

    yield
    # Clean up the ML models and release the resources
    ml_models.clear()


app = FastAPI(lifespan=lifespan)


@app.get("/predict")
async def predict(x: float):
    result = ml_models["answer_to_everything"](x)
    return {"result": result}
```

**Note**: This is the recommended way going forward, instead of using `startup` and `shutdown` events.

negotiations. The Indemnified Contributor may participate in any such claim at its own expense. For example, a Contributor might include the Program in a commercial product offering, Product X. That Contributor is then a Commercial Contributor. If that Commercial Contributor then makes performance claims, or offers warranties related to Product X, those performance claims and warranties are such Commercial Contributor's responsibility alone. Under this section, the Commercial Contributor would have to...

* Improve scheduler performance of MatchInterPodAffinity predicate. ([#57476](https://github.com/kubernetes/kubernetes/pull/57476), [@misterikkit](https://github.com/misterikkit))
* Improve scheduler performance of MatchInterPodAffinity predicate. ([#57477](https://github.com/kubernetes/kubernetes/pull/57477), [@misterikkit](https://github.com/misterikkit))

    *   The VolumeScheduling feature gate must also be enabled on kube-scheduler and kube-controller-manager components.
    *   The NoVolumeNodeConflict predicate has been removed.  For non-default schedulers, update your scheduler policy.
    *   The CheckVolumeBinding predicate must be enabled in non-default schedulers. ([#55039](https://github.com/kubernetes/kubernetes/pull/55039),[ @msau42](https://github.com/msau42))

### **OpenStack**

* Performance of the affinity/anti-affinity predicate for the default scheduler has been significantly improved. ([#62211](https://github.com/kubernetes/kubernetes/pull/62211), [@bsalamat](https://github.com/bsalamat))

- The `Should recreate evicted statefulset` conformance [test]( https://github.com/kubernetes/kubernetes/blob/master/test/e2e/apps/statefulset.go) fails because `Pod ss-0 expected to be re-created at least once`. This was caused by the `Predicate PodFitsHostPorts failed` scheduling error. The root cause was a host port conflict for port `21017`. This port was in-use as an ephemeral port by another application running on the node. This will be looked at for the 1.18 release.

- Performance of the anti-affinity predicate of the default scheduler has been improved. ([#66948](https://github.com/kubernetes/kubernetes/pull/66948), [@mohamed-mehany](https://github.com/mohamed-mehany))

      - `scheduler_scheduling_algorithm_latency_microseconds` -> `scheduler_scheduling_algorithm_duration_seconds`
      - `scheduler_scheduling_algorithm_predicate_evaluation` -> `scheduler_scheduling_algorithm_predicate_evaluation_seconds`
      - `scheduler_scheduling_algorithm_priority_evaluation` -> `scheduler_scheduling_algorithm_priority_evaluation_seconds`

### Bug or Regression

- Fix 121094 by re-introducing the readiness predicate for externalTrafficPolicy: Local services. ([#121116](https://github.com/kubernetes/kubernetes/pull/121116), [@alexanderConstantinescu](https://github.com/alexanderConstantinescu)) [SIG Cloud Provider and Network]