在 S3存储桶之间移动文件的最佳方式？

Update

As pointed out by alberge (+1), nowadays the excellent AWS Command Line Interface provides the most versatile approach for interacting with (almost) all things AWS - it meanwhile covers most services' APIs and also features higher level S3 commands for dealing with your use case specifically, see the AWS CLI reference for S3:

• sync - Syncs directories and S3 prefixes. Your use case is covered by Example 2 (more fine grained usage with --exclude, --include and prefix handling etc. is also available):

  The following sync command syncs objects under a specified prefix and bucket to objects under another specified prefix and bucket by copying s3 objects. [...]

  aws s3 sync s3://from_my_bucket s3://to_my_other_bucket
  

For completeness, I'll mention that the lower level S3 commands are also still available via the s3api sub command, which would allow to directly translate any SDK based solution to the AWS CLI before adopting its higher level functionality eventually.

Initial Answer

Moving files between S3 buckets can be achieved by means of the PUT Object - Copy API (followed by DELETE Object):

This implementation of the PUT operation creates a copy of an object that is already stored in Amazon S3. A PUT copy operation is the same as performing a GET and then a PUT. Adding the request header, x-amz-copy-source, makes the PUT operation copy the source object into the destination bucket. Source

There are respective samples for all existing AWS SDKs available, see Copying Objects in a Single Operation. Naturally, a scripting based solution would be the obvious first choice here, so Copy an Object Using the AWS SDK for Ruby might be a good starting point; if you prefer Python instead, the same can be achieved via boto as well of course, see method copy_key() within boto's S3 API documentation.

PUT Object only copies files, so you'll need to explicitly delete a file via DELETE Object still after a successful copy operation, but that will be just another few lines once the overall script handling the bucket and file names is in place (there are respective examples as well, see e.g. Deleting One Object Per Request).

If you have a unix host within AWS, then use s3cmd from s3tools.org. Set up permissions so that your key as read access to your development bucket. Then run:

s3cmd cp -r s3://productionbucket/feed/feedname/date s3://developmentbucket/feed/feedname

We had this exact problem with our ETL jobs at Snowplow, so we extracted our parallel file-copy code (Ruby, built on top of Fog), into its own Ruby gem, called Sluice:

https://github.com/snowplow/sluice

Sluice also handles S3 file delete, move and download; all parallelised and with automatic re-try if an operation fails (which it does surprisingly often). I hope it's useful!

Here is a ruby class for performing this: https://gist.github.com/4080793

Example usage:

$ gem install aws-sdk
$ irb -r ./bucket_sync_service.rb
> from_creds = {aws_access_key_id:"XXX",
aws_secret_access_key:"YYY",
bucket:"first-bucket"}
> to_creds = {aws_access_key_id:"ZZZ",
aws_secret_access_key:"AAA",
bucket:"first-bucket"}
> syncer = BucketSyncService.new(from_creds, to_creds)
> syncer.debug = true # log each object
> syncer.perform

.NET Example as requested:

using (client)
{
var existingObject = client.ListObjects(requestForExisingFile).S3Objects;
if (existingObject.Count == 1)
{
var requestCopyObject = new CopyObjectRequest()
{
SourceBucket = BucketNameProd,
SourceKey = objectToMerge.Key,
DestinationBucket = BucketNameDev,
DestinationKey = newKey
};
client.CopyObject(requestCopyObject);
}
}

with client being something like

var config = new AmazonS3Config { CommunicationProtocol = Protocol.HTTP, ServiceURL = "s3-eu-west-1.amazonaws.com" };
var client = AWSClientFactory.CreateAmazonS3Client(AWSAccessKey, AWSSecretAccessKey, config);

There might be a better way, but it's just some quick code I wrote to get some files transferred.

To move/copy from one bucket to another or the same bucket I use s3cmd tool and works fine. For instance:

s3cmd cp --recursive s3://bucket1/directory1 s3://bucket2/directory1
s3cmd mv --recursive s3://bucket1/directory1 s3://bucket2/directory1

The new official AWS CLI natively supports most of the functionality of s3cmd. I'd previously been using s3cmd or the ruby AWS SDK to do things like this, but the official CLI works great for this.

http://docs.aws.amazon.com/cli/latest/reference/s3/sync.html

aws s3 sync s3://oldbucket s3://newbucket

I know this is an old thread but for others who reach there my suggestion is to create a scheduled job to copy content from production bucket to development one.

You can use If you use .NET this article might help you

https://edunyte.com/2015/03/aws-s3-copy-object-from-one-bucket-or/

For me the following command just worked:

aws s3 mv s3://bucket/data s3://bucket/old_data --recursive

Actually as of recently I just use the copy+paste action in the AWS s3 interface. Just navigate to the files you want to copy, click on "Actions" -> "Copy" then navigate to the destination bucket and "Actions" -> "Paste"

It transfers the files pretty quick and it seems like a less convoluted solution that doesn't require any programming, or over the top solutions like that.

I spent days writing my own custom tool to parallelize the copies required for this, but then I ran across documentation on how to get the AWS S3 CLI sync command to synchronize buckets with massive parallelization. The following commands will tell the AWS CLI to use 1,000 threads to execute jobs (each a small file or one part of a multipart copy) and look ahead 100,000 jobs:

aws configure set default.s3.max_concurrent_requests 1000
aws configure set default.s3.max_queue_size 100000

After running these, you can use the simple sync command as follows:

aws s3 sync s3://source-bucket/source-path s3://destination-bucket/destination-path

On an m4.xlarge machine (in AWS--4 cores, 16GB RAM), for my case (3-50GB files) the sync/copy speed went from about 9.5MiB/s to 700+MiB/s, a speed increase of 70x over the default configuration.

Update: Note that S3CMD has been updated over the years and these changes are now only effective when you're working with lots of small files. Also note that S3CMD on Windows (only on Windows) is seriously limited in overall throughput and can only achieve about 3Gbps per process no matter what instance size or settings you use. Other systems like S5CMD have the same problem. I've spoken to the S3 team about this and they're looking into it.

For new version aws2.

aws2 s3 sync s3://SOURCE_BUCKET_NAME s3://NEW_BUCKET_NAME

If you're working in Python you can use cloudpathlib, which wraps boto3 to copy from one bucket to another.

Because it uses the AWS copy operation when going from an S3 source to an S3 target, it doesn't actually download and then re-upload any data—just asks AWS to move the file to the new location.

First, be sure to be authenticated properly with an ~/.aws/credentials file or environment variables set with an account that can access both buckets. See more options in the cloudpathlib docs.

Here's how you could copy files from one bucket to another:

from cloudpathlib import CloudPath


source = CloudPath("s3://bucket1/source.txt")
destination = CloudPath("s3://bucket2/destination.txt")


# create the source file
source.write_text("hello!")


# destination does not exist
destination.exists()
#> True


# move the source file
source.copy(destination)
#> S3Path('s3://bucket2/destination.txt')


# destination now exists
destination.exists()
#> True


# it has the expected content
destination.read_text()
#> 'hello!'

I would like to add some stuff to this question.

I got an issue while uploading a .gz file with 75GB.

The error was

An error occurred (InvalidArgument) when calling the UploadPart operation: Part number must be an integer between 1 and 10000, inclusive

I did the below changes

aws configure set default.s3.multipart_chunksize 64MB
aws configure set default.s3.max_concurrent_requests 1000
aws configure set default.s3.max_queue_size 100000

In addition to that, I contact customer support to ensure it'll work fine in future uploads.

Here is the response from AWS customer support.

To mitigate this issue, we have modified the AWS CLI S3 Configuration by executing the below "aws configure set" command to increase the multipart_chunksize value for the default profile:

           $ aws configure set default.s3.multipart_chunksize 64MB

AWS CLI S3 Configuration : https://docs.aws.amazon.com/cli/latest/topic/s3-config.html

Please be informed that the parameter --expected-size (string) is to specify the expected size of a stream in terms of bytes. This argument is needed only when a stream is being uploaded to s3 and the size is larger than 50GB. Failure to include this argument under these conditions may result in a failed upload due to too many parts in upload.

Hence, I would also request you to please provide/pass the parameter --expected-size along with the command that you are executing. For example, please refer the below 'aws s3 cp' command:

           $ aws s3 cp - s3://mybucket/stream.txt --expected-size 54760833024

'aws s3 cp' CLI Reference : https://awscli.amazonaws.com/v2/documentation/api/latest/reference/s3/cp.html

AWS now also allows replication for S3 buckets: https://aws.amazon.com/s3/features/replication/

You can set up filters if you want to copy specific type of files: https://docs.aws.amazon.com/AmazonS3/latest/userguide/replication-add-config.html#replication-config-optional-filter

Additionally you can use batch replication if it is a one time activity: https://aws.amazon.com/blogs/aws/new-replicate-existing-objects-with-amazon-s3-batch-replication/

Please note that live replication does not copy existing objects. So you'll need to use a combination of batch replication + live replication to sync your S3 buckets.