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::HTTP - use HTTP with IO::Async


   use Future::AsyncAwait;

   use IO::Async::Loop;
   use Net::Async::HTTP;
   use URI;

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

   my $http = Net::Async::HTTP->new();

   $loop->add( $http );

   my $response = await $http->do_request(
      uri => URI->new( "" ),

   print "Front page of is:\n";
   print $response->as_string;


This object class implements an asynchronous HTTP user agent. It sends requests to servers, returning Future instances to yield responses when they are received. The object supports multiple concurrent connections to servers, and allows multiple requests in the pipeline to any one connection. Normally, only one such object will be needed per program to support any number of requests.

As well as using futures the module also supports a callback-based interface.

This module optionally supports SSL connections, if IO::Async::SSL is installed. If so, SSL can be requested either by passing a URI with the https scheme, or by passing a true value as the SSL parameter.

Connection Pooling

There are three ways in which connections to HTTP server hosts are managed by this object, controlled by the value of max_connections_per_host. This controls when new connections are established to servers, as compared to waiting for existing connections to be free, as new requests are made to them.

They are:

max_connections_per_host = 1

This is the default setting. In this mode, there will be one connection per host on which there are active or pending requests. If new requests are made while an existing one is outstanding, they will be queued to wait for it.

If pipelining is active on the connection (because both the pipeline option is true and the connection is known to be an HTTP/1.1 server), then requests will be pipelined into the connection awaiting their response. If not, they will be queued awaiting a response to the previous before sending the next.

max_connections_per_host > 1

In this mode, there can be more than one connection per host. If a new request is made, it will try to re-use idle connections if there are any, or if they are all busy it will create a new connection to the host, up to the configured limit.

max_connections_per_host = 0

In this mode, there is no upper limit to the number of connections per host. Every new request will try to reuse an idle connection, or else create a new one if all the existing ones are busy.

These modes all apply per hostname / server port pair; they do not affect the behaviour of connections made to differing hostnames, or differing ports on the same hostname.


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

user_agent => STRING

A string to set in the User-Agent HTTP header. If not supplied, one will be constructed that declares Net::Async::HTTP and the version number.

headers => ARRAY or HASH

Since version 0.45.

A set of extra headers to apply to every outgoing request. May be specified either as an even-sized array containing key/value pairs, or a hash.

Individual header values may be added or changed without replacing the entire set by using the configure method and passing a key called +headers:

   $http->configure( +headers => { One_More => "Key" } );

max_redirects => INT

Optional. How many levels of redirection to follow. If not supplied, will default to 3. Give 0 to disable redirection entirely.

max_in_flight => INT

Optional. The maximum number of in-flight requests to allow per host when pipelining is enabled and supported on that host. If more requests are made over this limit they will be queued internally by the object and not sent to the server until responses are received. If not supplied, will default to 4. Give 0 to disable the limit entirely.

max_connections_per_host => INT

Optional. Controls the maximum number of connections per hostname/server port pair, before requests will be queued awaiting one to be free. Give 0 to disable the limit entirely. See also the "Connection Pooling" section documented above.

Currently, if not supplied it will default to 1. However, it has been found in practice that most programs will raise this limit to something higher, perhaps 3 or 4. Therefore, a future version of this module may set a higher value.

To test if your application will handle this correctly, you can set a different default by setting an environment variable:


timeout => NUM

Optional. How long in seconds to wait before giving up on a request. If not supplied then no default will be applied, and no timeout will take place.

stall_timeout => NUM

Optional. How long in seconds to wait after each write or read of data on a socket, before giving up on a request. This may be more useful than timeout on large-file operations, as it will not time out provided that regular progress is still being made.

proxy_host => STRING

proxy_port => INT

Since version 0.10.

proxy_path => PATH

Since version 0.49.

Optional. Default values to apply to each request method.

Optional. A reference to a HTTP::Cookies object. Will be used to set cookies in requests and store them from responses.

pipeline => BOOL

Optional. If false, disables HTTP/1.1-style request pipelining.

close_after_request => BOOL

Since version 0.45.

Optional. If true, will set the Connection: close header on outgoing requests and disable pipelining, thus making every request use a new connection.

family => INT

local_host => STRING

local_port => INT

local_addrs => ARRAY

local_addr => HASH or ARRAY

Optional. Parameters to pass on to the connect method used to connect sockets to HTTP servers. Sets the socket family and local socket address to bind() to. For more detail, see the documentation in IO::Async::Connector.

fail_on_error => BOOL

Optional. Affects the behaviour of response handling when a 4xx or 5xx response code is received. When false, these responses will be processed as other responses and yielded as the result of the future, or passed to the on_response callback. When true, such an error response causes the future to fail, or the on_error callback to be invoked.

The HTTP response and request objects will be passed as well as the code and message, and the failure name will be http.

   ( $code_message, "http", $response, $request ) = $f->failure

   $on_error->( "$code $message", $response, $request )

read_len => INT

write_len => INT

Optional. Used to set the reading and writing buffer lengths on the underlying IO::Async::Stream objects that represent connections to the server. If not define, a default of 64 KiB will be used.

ip_tos => INT or STRING

Optional. Used to set the IP_TOS socket option on client sockets. If given, should either be a IPTOS_* constant, or one of the string names lowdelay, throughput, reliability or mincost. If undefined or left absent, no option will be set.

decode_content => BOOL

Optional. If true, incoming responses that have a recognised Content-Encoding are handled by the module, and decompressed content is passed to the body handling callback or returned in the HTTP::Response. See "CONTENT DECODING" below for details of which encoding types are recognised. When this option is enabled, outgoing requests also have the Accept-Encoding header added to them if it does not already exist.

Currently the default is false, because this behaviour is new, but it may default to true in a later version. Applications which care which behaviour applies should set this to a defined value to ensure it doesn't change.


Additionally, any parameters whose names start with SSL_ will be stored and passed on requests to perform SSL requests. This simplifies configuration of common SSL parameters.

require_SSL => BOOL

Optional. If true, then any attempt to make a request that does not use SSL (either by calling request, or as a result of a redirection) will immediately fail.


Since version 0.42.

Additionally, any parameters whose names start with SOCKS_ will be stored and used by Net::Async::SOCKS to establish connections via a configured proxy.


The following methods documented in an await expression return Future instances.

When returning a Future, the following methods all indicate HTTP-level errors using the Future failure name of http. If the error relates to a specific response it will be included. The original request is also included.

   $f->fail( $message, "http", $response, $request )


   $response = await $http->do_request( %args );

Send an HTTP request to a server, returning a Future that will yield the response. The request may be represented by an HTTP::Request object, or a URI object, depending on the arguments passed.

The following named arguments are used for HTTP::Requests:

request => HTTP::Request

A reference to an HTTP::Request object

host => STRING

Hostname of the server to connect to

port => INT or STRING

Optional. Port number or service of the server to connect to. If not defined, will default to http or https depending on whether SSL is being used.

family => INT or STRING

Optional. Restricts the socket family for connecting. If not defined, will default to the globally-configured value in the object. The value may either be a PF_* constant directly, or the lowercase name of one such as inet.


Optional. If true, an SSL connection will be used.

The following named arguments are used for URI requests:

uri => URI or STRING

A reference to a URI object, or a plain string giving the request URI. If the scheme is https then an SSL connection will be used.

method => STRING

Optional. The HTTP method name. If missing, GET is used.

content => STRING or ARRAY ref

Optional. The body content to use for PUT or POST requests.

If this is a plain scalar it will be used directly, and a content_type field must also be supplied to describe it.

If this is an ARRAY ref and the request method is POST, it will be form encoded. It should contain an even-sized list of field names and values. For more detail see "POST" in HTTP::Request::Common.

content_type => STRING

The type of non-form data content.

user => STRING
pass => STRING

Optional. If both are given, the HTTP Basic Authorization header will be sent with these details.

headers => ARRAY|HASH

Optional. If provided, contains additional HTTP headers to set on the constructed request object. If provided as an ARRAY reference, it should contain an even-sized list of name/value pairs.

proxy_host => STRING
proxy_port => INT

Since version 0.10.

Optional. Override the hostname or port number implied by the URI.

proxy_path => PATH

Since version 0.49.

Optional. Set a UNIX socket path to use as a proxy. To make use of this, also set the family argument to unix.

For either request type, it takes the following arguments:

request_body => STRING | CODE | Future

Optional. Allows request body content to be generated by a future or callback, rather than being provided as part of the request object. This can either be a plain string, a CODE reference to a generator function, or a future.

As this is passed to the underlying IO::Async::Stream write method, the usual semantics apply here. If passed a CODE reference, it will be called repeatedly whenever it's safe to write. The code should should return undef to indicate completion. If passed a Future it is expected to eventually yield the body value.

As with the content parameter, the content_type field should be specified explicitly in the request header, as should the content length (typically via the HTTP::Request content_length method). See also examples/

expect_continue => BOOL

Optional. If true, sets the Expect request header to the value 100-continue and does not send the request_body parameter until a 100 Continue response is received from the server. If an error response is received then the request_body code, if present, will not be invoked.

on_ready => CODE

Optional. A callback that is invoked once a socket connection is established with the HTTP server, but before the request is actually sent over it. This may be used by the client code to inspect the socket, or perform any other operations on it. This code is expected to return a Future; only once that has completed will the request cycle continue. If it fails, that failure is propagated to the caller.

   $f = $on_ready->( $connection )
on_redirect => CODE

Optional. A callback that is invoked if a redirect response is received, before the new location is fetched. It will be passed the response and the new URL.

   $on_redirect->( $response, $location )
on_body_write => CODE

Optional. A callback that is invoked after each successful syswrite of the body content. This may be used to implement an upload progress indicator or similar. It will be passed the total number of bytes of body content written so far (i.e. excluding bytes consumed in the header).

   $on_body_write->( $written )
max_redirects => INT

Optional. How many levels of redirection to follow. If not supplied, will default to the value given in the constructor.

timeout => NUM
stall_timeout => NUM

Optional. Overrides the object's configured timeout values for this one request. If not specified, will use the configured defaults.

On a timeout, the returned future will fail with either timeout or stall_timeout as the operation name.

   ( $message, "timeout" ) = $f->failure

do_request (void)

   $http->do_request( %args )

When not returning a future, the following extra arguments are used as callbacks instead:

on_response => CODE

A callback that is invoked when a response to this request has been received. It will be passed an HTTP::Response object containing the response the server sent.

   $on_response->( $response )
on_header => CODE

Alternative to on_response. A callback that is invoked when the header of a response has been received. It is expected to return a CODE reference for handling chunks of body content. This CODE reference will be invoked with no arguments once the end of the request has been reached, and whatever it returns will be used as the result of the returned Future, if there is one.

   $on_body_chunk = $on_header->( $header )

      $on_body_chunk->( $data )
      $response = $on_body_chunk->()
on_error => CODE

A callback that is invoked if an error occurs while trying to send the request or obtain the response. It will be passed an error message.

   $on_error->( $message )

If this is invoked because of a received 4xx or 5xx error code in an HTTP response, it will be invoked with the response and request objects as well.

   $on_error->( $message, $response, $request )


   $response = await $http->GET( $uri, %args );

   $response = await $http->HEAD( $uri, %args );

   $response = await $http->PUT( $uri, $content, %args );

   $response = await $http->POST( $uri, $content, %args );

Since version 0.36.

   $response = await $http->PATCH( $uri, $content, %args );

Since version 0.48.

   $response = await $http->DELETE( $uri, %args );

Since version 0.49.

Convenient wrappers for performing GET, HEAD, PUT, POST, PATCH or DELETE requests with a URI object and few if any other arguments, returning a Future.

Remember that POST with non-form data (as indicated by a plain scalar instead of an ARRAY reference of form data name/value pairs) needs a content_type key in %args.


The following methods are intended as points for subclasses to override, to add extra functionallity.


   $http->prepare_request( $request )

Called just before the HTTP::Request object is sent to the server.


   $http->process_response( $response )

Called after a non-redirect HTTP::Response has been received from a server. The originating request will be set in the object.


If the required decompression modules are installed and available, compressed content can be decoded. If the received Content-Encoding is recognised and the required module is available, the content is transparently decoded and the decoded content is returned in the resulting response object, or passed to the data chunk handler. In this case, the original Content-Encoding header will be deleted from the response, and its value will be available instead as X-Original-Content-Encoding.

The following content encoding types are recognised by these modules:

  • gzip (q=0.7) and deflate (q=0.5)

    Recognised if Compress::Raw::Zlib version 2.057 or newer is installed.

  • bzip2 (q=0.8)

    Recognised if Compress::Bzip2 version 2.10 or newer is installed.

Other content encoding types can be registered by calling the following method


   Net::Async::HTTP->register_decoder( $name, $q, $make_decoder )

Registers an encoding type called $name, at the quality value $q. In order to decode this encoding type, $make_decoder will be invoked with no paramters, and expected to return a CODE reference to perform one instance of decoding.

   $decoder = $make_decoder->()

This decoder will be invoked on string buffers to decode them until the end of stream is reached, when it will be invoked with no arguments.

   $content = $decoder->( $encoded_content )
   $content = $decoder->() # EOS


Concurrent GET

The Future-returning GET method makes it easy to await multiple URLs at once, by using the Future::Utils fmap_void utility

   use Future::AsyncAwait;
   use Future::Utils qw( fmap_void );

   my @URLs = ( ... );

   my $http = Net::Async::HTTP->new( ... );
   $loop->add( $http );

   my $future = fmap_void {
      my ( $url ) = @_;
      $http->GET( $url )
           ->on_done( sub {
              my $response = shift;
              say "$url succeeded: ", $response->code;
              say "  Content-Type:", $response->content_type;
           } )
           ->on_fail( sub {
              my $failure = shift;
              say "$url failed: $failure";
           } );
   } foreach => \@URLs,
     concurrent => 5;

   await $future;



Parts of this code, or bugfixes to it were paid for by


Paul Evans <>