The London Perl and Raku Workshop takes place on 26th Oct 2024. If your company depends on Perl, please consider sponsoring and/or attending.


Net::Async::Webservice::S3 - use Amazon's S3 web service with IO::Async


   use IO::Async::Loop;
   use Net::Async::Webservice::S3;

   my $loop = IO::Async::Loop->new;

   my $s3 = Net::Async::Webservice::S3->new(
      access_key => ...,
      secret_key => ...,
      bucket     => "my-bucket-here",
   $loop->add( $s3 );

   my $put_f = $s3->put_object(
      key   => "the-key",
      value => "A new value for the key\n";

   my $get_f = $s3->get_object(
      key => "another-key",

   $loop->await_all( $put_f, $get_f );

   print "The value is:\n", $get_f->get;


This module provides a webservice API around Amazon's S3 web service for use in an IO::Async-based program. Each S3 operation is represented by a method that returns a Future; this future, if successful, will eventually return the result of the operation.


The following named parameters may be passed to new or configure:

http => Net::Async::HTTP

Optional. Allows the caller to provide a specific asynchronous HTTP user agent object to use. This will be invoked with a single method, as documented by Net::Async::HTTP:

   $response_f = $http->do_request( request => $request, ... )

If absent, a new instance will be created and added as a child notifier of this object. If a value is supplied, it will be used as supplied and not specifically added as a child notifier. In this case, the caller must ensure it gets added to the underlying IO::Async::Loop instance, if required.

access_key => STRING

secret_key => STRING

The twenty-character Access Key ID and forty-character Secret Key to use for authenticating requests to S3.

ssl => BOOL

Optional. If given a true value, will use https URLs over SSL.

This setting defaults on, but can be disabled by passing a defined-but-false value (such as 0).

bucket => STRING

Optional. If supplied, gives the default bucket name to use, at which point it is optional to supply to the remaining methods.

prefix => STRING

Optional. If supplied, this prefix string is prepended to any key names passed in methods, and stripped from the response from list_bucket. It can be used to keep operations of the object contained within the named key space. If this string is supplied, don't forget that it should end with the path delimiter in use by the key naming scheme (for example /).

host => STRING

Optional. Sets the hostname to talk to the S3 service. Usually the default of is sufficient. This setting allows for communication with other service providers who provide the same API as S3.

max_retries => INT

Optional. Maximum number of times to retry a failed operation. Defaults to 3.

list_max_keys => INT

Optional. Maximum number of keys at a time to request from S3 for the list_bucket method. Larger values may be more efficient as fewer roundtrips will be required per method call. Defaults to 1000.

part_size => INT

Optional. Size in bytes to break content for using multipart upload. If an object key's size is no larger than this value, multipart upload will not be used. Defaults to 100 MiB.

read_size => INT

Optional. Size in bytes to read per call to the $gen_value content generation function in put_object. Defaults to 64 KiB. Be aware that too large a value may lead to the PUT stall timer failing to be invoked on slow enough connections, causing spurious timeouts.

timeout => NUM

Optional. If configured, this is passed into individual requests of the underlying Net::Async::HTTP object, except for the actual content GET or PUT operations. It is therefore used by list_bucket, delete_object, and the multi-part metadata operations used by put_object. To apply an overall timeout to an individual get_object or put_object operation, pass a specific timeout argument to those methods specifically.

stall_timeout => NUM

Optional. If configured, this is passed into the underlying Net::Async::HTTP object and used for all content uploads and downloads.

put_concurrent => INT

Optional. If configured, gives a default value for the concurrent parameter to put_object.


The following methods all support the following common arguments:

timeout => NUM
stall_timeout => NUM

Optional. Passed directly to the underlying Net::Async::HTTP->request method.

The following methods documented with a trailing call to ->get return Future instances.


   ( $keys, $prefixes ) = $s3->list_bucket( %args )->get

Requests a list of the keys in a bucket, optionally within some prefix.

Takes the following named arguments:

bucket => STR

The name of the S3 bucket to query

prefix => STR
delimiter => STR

Optional. If supplied, the prefix and delimiter to use to divide up the key namespace. Keys will be divided on the delimiter parameter, and only the key space beginning with the given prefix will be queried.

The Future will return two ARRAY references. The first provides a list of the keys found within the given prefix, and the second will return a list of the common prefixes of further nested keys.

Each key in the $keys list is given in a HASH reference containing

key => STRING

The key's name

last_modified => STRING

The last modification time of the key given in ISO 8601 format

etag => STRING

The entity tag of the key

size => INT

The size of the key's value, in bytes

storage_class => STRING

The S3 storage class of the key

Each key in the $prefixes list is given as a plain string.


   ( $value, $response, $meta ) = $s3->get_object( %args )

Requests the value of a key from a bucket.

Takes the following named arguments:

bucket => STR

The name of the S3 bucket to query

key => STR

The name of the key to query

on_chunk => CODE

Optional. If supplied, this code will be invoked repeatedly on receipt of more bytes of the key's value. It will be passed the HTTP::Response object received in reply to the request, and a byte string containing more bytes of the value. Its return value is not important.

   $on_chunk->( $header, $bytes )

If this is supplied then the key's value will not be accumulated, and the final result of the Future will be an empty string.

byte_range => STRING

Optional. If supplied, is used to set the Range request header with bytes as the units. This gives a range of bytes of the object to fetch, rather than fetching the entire content. The value must be as specified by HTTP/1.1; i.e. a comma-separated list of ranges, where each range specifies a start and optionally an inclusive stop byte index, separated by hypens.

if_match => STRING

Optional. If supplied, is used to set the If-Match request header to the given string, which should be an entity etag. If the requested object no longer has this etag, the request will fail with an http failure whose response code is 412.

The Future will return a byte string containing the key's value, the HTTP::Response that was received, and a hash reference containing any of the metadata fields, if found in the response. If an on_chunk code reference is passed, the $value string will be empty.

If the entire content of the object is requested (i.e. if byte_range is not supplied) then stall timeout failures will be handled specially. If a stall timeout happens while receiving the content, the request will be retried using the Range header to resume from progress so far. This will be repeated while every attempt still makes progress, and such resumes will not be counted as part of the normal retry count. The resume request also uses If-Match to ensure it only resumes the resource with matching ETag. If a resume request fails for some reason (either because the ETag no longer matches or something else) then this error is ignored, and the original stall timeout failure is returned.


   ( $response, $meta ) = $s3->head_object( %args )->get

Requests the value metadata of a key from a bucket. This is similar to the get_object method, but uses the HEAD HTTP verb instead of GET.

Takes the same named arguments as get_object, but will ignore an on_chunk callback, if provided.

The Future will return the HTTP::Response object and metadata hash reference, without the content string (as no content is returned to a HEAD request).


   ( $value_f, $response, $meta ) = $s3->head_then_get_object( %args )->get

   ( $value, $response, $meta ) = $value_f->get

Performs a GET operation similar to get_object, but allows access to the metadata header before the body content is complete.

Takes the same named arguments as get_object.

The returned Future completes as soon as the metadata header has been received and yields a second future (the body future), the HTTP::Response and a hash reference containing the metadata fields. The body future will eventually yield the actual body, along with another copy of the response and metadata hash reference.


   ( $etag, $length ) = $s3->put_object( %args ) ==> ( $etag, $length )

Sets a new value for a key in the bucket.

Takes the following named arguments:

bucket => STRING

The name of the S3 bucket to put the value in

key => STRING

The name of the key to put the value in

value => STRING
value => Future giving STRING

Optional. If provided, gives a byte string as the new value for the key or a Future which will eventually yield such.

value => CODE
value_length => INT

Alternative form of value, which is a CODE reference to a generator function. It will be called repeatedly to generate small chunks of content, being passed the position and length it should yield.

   $chunk = $value->( $pos, $len )

Typically this can be provided either by a substr operation on a larger string buffer, or a sysseek and sysread operation on a filehandle.

In normal operation the function will just be called in a single sweep in contiguous regions up to the extent given by value_length. If however, the MD5sum check fails at the end of upload, it will be called again to retry the operation. The function must therefore be prepared to be invoked multiple times over its range.

value => Future giving ( CODE, INT )

Alternative form of value, in which a Future eventually yields the value generation CODE reference and length. The CODE reference is invoked as documented above.

   ( $gen_value, $value_len ) = $value->get;

   $chunk = $gen_value->( $pos, $len );
gen_parts => CODE

Alternative to value in the case of larger values, and implies the use of multipart upload. Called repeatedly to generate successive parts of the upload. Each time gen_parts is called it should return one of the forms of value given above; namely, a byte string, a CODE reference and size pair, or a Future which will eventually yield either of these forms.

   ( $value ) = $gen_parts->()

   ( $gen_value, $value_length ) = $gen_parts->()

   ( $value_f ) = $gen_parts->(); $value = $value_f->get
                                  ( $gen_value, $value_length ) = $value_f->get

Each case is analogous to the types that the value key can take.

meta => HASH

Optional. If provided, gives additional user metadata fields to set on the object, using the X-Amz-Meta- fields.

timeout => NUM

Optional. For single-part uploads, this sets the timeout argument to use for the actual PUT request. For multi-part uploads, this argument is currently ignored.

meta_timeout => NUM

Optional. For multipart uploads, this sets the timeout argument to use for the initiate and complete requests, overriding a configured timeout. Ignored for single-part uploads.

part_timeout => NUM

Optional. For multipart uploads, this sets the timeout argument to use for the individual part PUT requests. Ignored for single-part uploads.

on_write => CODE

Optional. If provided, this code will be invoked after each successful syswrite call on the underlying filehandle when writing actual file content, indicating that the data was at least written as far as the kernel. It will be passed the total byte length that has been written for this call to put_object. By the time the call has completed, this will be the total written length of the object.

   $on_write->( $bytes_written )

Note that because of retries it is possible this count will decrease, if a part has to be retried due to e.g. a failing MD5 checksum.

concurrent => INT

Optional. If present, gives the number of parts to upload concurrently. If absent, a default of 1 will apply (i.e. no concurrency).

The Future will return a string containing the S3 ETag of the newly-set key, and the length of the value in bytes.

For single-part uploads the ETag will be the MD5 sum in hex, surrounded by quote marks. For multi-part uploads this is a string in a different form, though details of its generation are not specified by S3.

The returned MD5 sum from S3 during upload will be checked against an internally-generated MD5 sum of the content that was sent, and an error result will be returned if these do not match.


   $s3->delete_object( %args )->get

Deletes a key from the bucket.

Takes the following named arguments:

bucket => STRING

The name of the S3 bucket to put the value in

key => STRING

The name of the key to put the value in

The Future will return nothing.


Parts of this code were paid for by


Paul Evans <>