When building web APIs with FastAPI, if there's an error in our code, FastAPI will normally contain it to the single request that triggered the error. 🛡

The client will get a **500 Internal Server Error** for that request, but the application will continue working for the next requests instead of just crashing completely.

### Bigger Errors - Crashes

Downloading and installing the package dependencies **could take minutes**, but using the **cache** would **take seconds** at most.

And as you would be building the container image again and again during development to check that your code changes are working, there's a lot of accumulated time this would save.

   * {@link MoreExecutors#directExecutor}. Normally, when the first future completes, all the other
   * tasks would be called recursively. Here, we detect that the delegate executor is executing
   * inline, and maintain a queue to dispatch tasks iteratively. There is one instance of this class
   * per call to submit() or submitAsync(), and each instance supports only one call to execute().
   *

<font color="#4E9A06">INFO</font>:     Application startup complete.
```

</div>

In the output, there's a line with something like:

```hl_lines="4"
INFO:     Uvicorn running on http://127.0.0.1:8000 (Press CTRL+C to quit)
```

That line shows the URL where your app is being served, in your local machine.

### Check it

* Check if there's a <a href="https://github.com/tiangolo/fastapi/discussions/categories/translations" class="external-link" target="_blank">GitHub Discussion</a> to coordinate translations for your language. You can subscribe to it, and when there's a new pull request to review, an automatic comment will be added to the discussion.

   * needed.
   *
   * A single spinlock ("busy") is used for initializing and
   * resizing the table, as well as populating slots with new Cells.
   * There is no need for a blocking lock; when the lock is not
   * available, threads try other slots (or the base).  During these
   * retries, there is increased contention and reduced locality,
   * which is still better than alternatives.
   *
   * Per-thread hash codes are initialized to random values.

message PodSchedulingContext {
  // Standard object metadata
  // +optional
  optional k8s.io.apimachinery.pkg.apis.meta.v1.ObjectMeta metadata = 1;

  // Spec describes where resources for the Pod are needed.
  optional PodSchedulingContextSpec spec = 2;

  // Status describes where resources for the Pod can be allocated.
  // +optional
  optional PodSchedulingContextStatus status = 3;
}

  cat <<EOF
There was a path found connecting //tensorflow/tools/pip_package:wheel
to a banned CUDA dependency when '--define framework_shared_object=false' is set.
This means that a CUDA target was probably included via an is_static condition,
used when targeting platforms like Windows where we build statically instead
of dynamically. Here's the output from bazel query:
EOF
  cat $BATS_TEST_TMPDIR/out

            //
            // TODO: This seems a little indirect. Is there a better way to do this? For FE1.0 there is
            // a special `TypeAliasConstructorDescriptor` for this case. For FIR there is
            // FirConstructor.originalConstructorIfTypeAlias but that doesn't seem to help here as it
            // is null for the constructors we get.

   * needed.
   *
   * A single spinlock ("busy") is used for initializing and
   * resizing the table, as well as populating slots with new Cells.
   * There is no need for a blocking lock; when the lock is not
   * available, threads try other slots (or the base).  During these
   * retries, there is increased contention and reduced locality,
   * which is still better than alternatives.
   *
   * Per-thread hash codes are initialized to random values.