# Load tensorflow-io-gcs-filesystem if enabled
if (_os.getenv("TF_USE_MODULAR_FILESYSTEM", "0") == "true" or
    _os.getenv("TF_USE_MODULAR_FILESYSTEM", "0") == "1"):
  import tensorflow_io_gcs_filesystem as _tensorflow_io_gcs_filesystem

# Lazy-load Keras v2/3.
_tf_uses_legacy_keras = (
    _os.environ.get("TF_USE_LEGACY_KERAS", None) in ("true", "True", "1"))
setattr(_current_module, "keras", _KerasLazyLoader(globals()))

By default, MinIO shards the objects across N/2 data and N/2 parity drives. Though, you can use [storage classes](https://github.com/minio/minio/tree/master/docs/erasure/storage-class) to use a custom configuration. We recommend N/2 data and parity blocks, as it ensures the best protection from drive failures.

MinIO server allows streaming compression to ensure efficient disk space usage.
Compression happens inflight, i.e objects are compressed before being written to disk(s).
MinIO uses [`klauspost/compress/s2`](https://github.com/klauspost/compress/tree/master/s2)
streaming compression due to its stability and performance.

This algorithm is specifically optimized for machine generated content.
Write throughput is typically at least 500MB/s per CPU core,

    /**
     * A phase in the lifecycle.
     *
     * A phase is identified by its name. It also contains a list of plugins bound to that phase,
     * a list of {@link Link links}, and a list of sub-phases.  This forms a tree of phases.
     */
    interface Phase {

        // ======================
        // Maven defined phases
        // ======================
        String ALL = "all";
        String EACH = "each";

			if err := adb("exec-out", "mkdir", "-p", path.Dir(deviceModDir)); err != nil {
				return 0, err
			}
			// We use a single recursive 'adb push' of the module root instead of
			// walking the tree and copying it piecewise. If the directory tree
			// contains nested modules this could push a lot of unnecessary contents,
			// but for the golang.org/x repos it seems to be significantly (~2x)

## Dependencies

### Added
- github.com/a8m/tree: [10a5fd5](https://github.com/a8m/tree/tree/10a5fd5)
- github.com/dougm/pretty: [2ee9d74](https://github.com/dougm/pretty/tree/2ee9d74)
- github.com/rasky/go-xdr: [4930550](https://github.com/rasky/go-xdr/tree/4930550)
- github.com/vmware/vmw-guestinfo: [25eff15](https://github.com/vmware/vmw-guestinfo/tree/25eff15)
- sigs.k8s.io/kustomize/kustomize/v5: v5.0.1

    [the Android flavor](https://github.com/google/guava/wiki/Android). You can
    find the Android Guava source in the [`android` directory].

[`android` directory]: https://github.com/google/guava/tree/master/android

## Adding Guava to your build

Guava's Maven group ID is `com.google.guava`, and its artifact ID is `guava`.
Guava provides two different "flavors": one for use on a (Java 8+) JRE and one

- GZIP or BZIP2 - CSV and JSON files can be compressed using GZIP, BZIP2, [ZSTD](https://facebook.github.io/zstd/), and streaming formats of [LZ4](https://lz4.github.io/lz4/), [S2](https://github.com/klauspost/compress/tree/master/s2#s2-compression) and [SNAPPY](http://google.github.io/snappy/).
- Parquet API supports columnar compression for  using GZIP, Snappy, LZ4. Whole object compression is not supported for Parquet objects.

- For instructions on using KES for encrypting the MinIO backend, follow the [KMS Quick Start](https://github.com/minio/minio/tree/master/docs/kms). The SSE-S3 configuration setup also supports MinIO KMS backend encryption.

## FAQ

> Why is this change needed?

Before, there were two separate mechanisms - S3 objects got encrypted using a KMS,

// Copyright (c) 2015-2021 MinIO, Inc.
//
// This file is part of MinIO Object Storage stack
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU Affero General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful