This is a parallel S3 and local filesystem execution tool.

With a correctly configured Go installation:

go get -u github.com/peakgames/s5cmd

This will install s5cmd in your $GOPATH/bin directory.

$ ./s5cmd --help

Usage: ./s5cmd [OPTION]... [COMMAND [PARAMS...]]
 
  -cs int
        Multipart chunk size in MB for uploads (default 5, auto calculated)
  -f string
        Commands-file or - for stdin
  -numworkers int
        Number of worker goroutines. Negative numbers mean multiples of runtime.NumCPU (default 256)
  -r int
        Retry S3 operations N times before failing (default 10)
  -stats
        Always print stats
  -version
        Prints current version

To list available commands, run without arguments.

The most powerful feature of s5cmd is the commands file. Thousands of S3 and filesystem commands are declared in a file (or simply piped in from another process) and they are executed using multiple parallel workers. Since only one program is launched, thousands of unnecessary fork-exec calls are avoided. This way S3 execution times can reach a few thousand operations per second.

See also: Nested Commands

Single commands are also supported with the s5cmd [command [params]] syntax. If this syntax is used, two workers are launched by default.

There are three main commands: cp, mv and rm. Arguments can be either S3 urls, S3 wildcards, local file/directory references or local glob patterns.

• Copy, Download or Upload: cp [options] [src] [dst]
• Move, Download or Upload and then delete: mv [options] [src] [dst]
• Delete: rm [src]
• Count objects and determine total size: du [src]
• Arbitrary shell-execute - ! commands...
• Exit - exit [exitcode] (see Exit Code)

• S3 urls should be in the format s3://bucket/key
• cp and mv commands accept the -n (no-clobber) option to prevent overwriting existing files or objects.
• Uploading cp and mv commands accept the -rr and -ia options to store objects in reduced-redundancy and infrequent-access modes respectively.
• Batch cp and mv commands also accept the --parents option to create the dir structure in destination. Dir structure is created from the first wildcard onwards.
• The du command only takes S3 arguments (prefix or wildcard)

cp s3://from-bucket/from-key s3://to-bucket/[to-key] # Copy object in S3 
mv s3://from-bucket/from-key s3://to-bucket/[to-key] # Move object in S3
rm s3://del-bucket/del-key # Delete S3 object
cp /path/to/src/file /path/to/dest[/] # Copy local file
mv /path/to/src/file /path/to/dest[/] # Move local file
cp s3://from-bucket/from-key /path/to/dest[/] # Download from S3
rm /path/to/del # Delete local file or (empty) directory
ls # List buckets
ls s3://bucket[/prefix] # List objects in bucket
cp /path/to/src s3://to-bucket/to-key[/] # Upload to S3
cp /path/to/src/dir/ s3://to-bucket/to-prefix/ # Upload directory to S3
cp /path/to/src/*.go s3://to-bucket/to-prefix/ # Upload glob to S3

Multiple-level wildcards are supported in S3 operations. This is achieved by listing all S3 objects with the prefix up to the first wildcard, then filtering the results in-memory. ie. For batch-downloads, first a ls call is made, the results are then converted to separate commands and executed in parallel.

Batch API is used deleting multiple S3 objects, so up to 1000 S3 objects can be deleted with a single call.

ls s3://bucket/prefix/*/file*gz # Wild-list objects in bucket
cp s3://from-bucket/prefix/*/file*gz /path/to/dest/ # Wild-download from S3
mv s3://from-bucket/prefix/*/file*gz /path/to/dest/ # Wild-download from S3, followed by delete
rm s3://from-bucket/prefix/*/file*gz # Wild-delete S3 objects (Batch-API)

• Comments start with a space followed by #, as in " # This is a comment"
• Empty lines are also ok
• -numworkers -1 means use runtime.NumCPU goroutines. -2 means 2*runtime.NumCPU and so on.
• The S3 throttling error SlowDown is exponentially retried. "Retryable operations" as specified by the AWS SDK (currently RequestError and RequestError) are retried by the SDK.

Success and fail commands can be specified with && and || operators. As the parser is pretty simple, multiple-level nested commands (doing something based on a result of a result) are not supported.

If you want to move an object between s3 buckets and then delete a local file if successful, you can do this:

mv s3://source-bkt/key s3://dest-bkt/key && rm /path/to/key

This is also valid:

mv a b/ || ! touch could-not-move # This is a comment

This as well:

! touch a && ! touch a-touched || ! touch a-couldnotbetouched

S3 credentials can be provided in a variety of ways.

Provide full S3 credentials with the environment variables:

AWS_ACCESS_KEY_ID=YOURKEY AWS_SECRET_ACCESS_KEY=ohnosecret AWS_REGION=us-east-1 ./s5cmd ls

Use the $HOME/.aws/credentials file:

[default]
aws_access_key_id = YOURKEY
aws_secret_access_key = ohnosecret
region = us-east-1

Then run s5cmd:

./s5cmd ls

To use a different profile, set the AWS_PROFILE env var. For more options, see the AWS SDK Configuration page.

The general output is in the format:

DATE TIME Short-Msg Detailed-Msg

• Trivial messages start with #, like number of workers or exit code and statistics
• +OK for successful operations: +OK "! touch touch-this-file"
• -ERR for failed operations: -ERR "! touche": executable file not found in $PATH
• ?ErrorCode for AWS-related errors, which will be retried (?SlowDown, ?InternalError, etc)

Item output (used in ls) is slightly different: DATE TIME fields are omitted, and the short-msg is a single + character.

Shell output (used in !) does not modify the executed command's output. Both stdout and stderr are mirrored.

If failed jobs are present, process exits with code 127. This can be overridden with the command exit, though in that case finishing the job list is not guaranteed.