If you are using containers, for example with Docker or Kubernetes, I'll tell you more about that in the next chapter: [FastAPI in Containers - Docker](docker.md){.internal-link target=_blank}.

In particular, when running on **Kubernetes** you will probably **not** want to use workers and instead run **a single Uvicorn process per container**, but I'll tell you about it later in that chapter.

///

## Multiple Workers { #multiple-workers }

	"github.com/minio/pkg/v3/policy"
)

// http Header "x-amz-content-sha256" == "UNSIGNED-PAYLOAD" indicates that the
// client did not calculate sha256 of the payload.
const unsignedPayload = "UNSIGNED-PAYLOAD"

// http Header "x-amz-content-sha256" == "STREAMING-UNSIGNED-PAYLOAD-TRAILER" indicates that the
// client did not calculate sha256 of the payload and there is a trailer.

        val allDistribution = bucket("${prefix}TestAllDistribution", "Declare a all distribution to be used by tests - useful for testing the final distribution that is published")
        val docsDistribution = bucket("${prefix}TestDocsDistribution", "Declare a docs distribution to be used by tests - useful for testing the final distribution that is published")

# Note that numpy 2.1.0 does not support python 3.9
numpy >= 2.0.0, < 2.2.0
wheel ~= 0.41.2
h5py >= 3.11.0, < 3.15.0
lit ~= 17.0.2
opt_einsum == 3.3.0
astunparse == 1.6.3
dill == 0.3.7
astor == 0.7.1
typing_extensions ~= 4.14.1
gast == 0.4.0
termcolor == 2.3.0
wrapt == 1.16.0
tblib == 2.0.0
ml_dtypes >= 0.5.1, < 0.6.0
auditwheel >= 6.1.0
# Install tensorboard, and keras

   * scheduled. If the task is already in the queue, the earliest execution time is used.
   *
   * The target execution time is implemented on a best-effort basis. If another task in this queue
   * is running when that time is reached, that task is allowed to complete before this task is
   * started. Similarly the task will be delayed if the host lacks compute resources.
   *

            }
            if (o instanceof CacheKey that) {
                return Objects.equals(groupId, that.groupId)
                        && Objects.equals(artifactId, that.artifactId)
                        && Objects.equals(version, that.version)
                        && Objects.equals(dependencyArtifacts, that.dependencyArtifacts)
                        && Objects.equals(workspace, that.workspace)

    verifier in modern browsers.
 *  Fix: Improve `CONNECT` handling for misbehaving HTTP proxies.
 *  Fix: Don't retry requests that failed due to timeouts.
 *  Fix: Cache 302s and 308s that include appropriate response headers.
 *  Fix: Improve pooling of connections that use proxy selectors.
 *  Fix: Don't leak connections when using ALPN on the desktop.

 *
 * <p>Here's an example that tests a {@code finalize} method:
 *
 * {@snippet :
 * final CountDownLatch latch = new CountDownLatch(1);
 * Object x = new MyClass() {
 *   ...
 *   protected void finalize() { latch.countDown(); ... }
 * };
 * x = null;  // Hint to the JIT that x is stack-unreachable
 * GcFinalization.await(latch);
 * }
 *
 * <p>Here's an example that uses a user-defined finalization predicate:
 *

 * methods for more details.
 *
 * <p>Note that this implementation is not synchronized. If multiple threads access the same cell of
 * an {@code ArrayTable} concurrently and one of the threads modifies its value, there is no
 * guarantee that the new value will be fully visible to the other threads. To guarantee that
 * modifications are visible, synchronize access to the table. Unlike other {@code Table}

   | it to several places.
   |
   | That repository definition will have a unique id, so we can create a mirror reference for that
   | repository, to be used as an alternate download site. The mirror site will be the preferred
   | server for that repository.
   |-->
  <mirrors>
    <!-- mirror
     | Specifies a repository mirror site to use instead of a given repository. The repository that