AnyEvent::HTTP(3) simple but non-blocking HTTP/HTTPS client


use AnyEvent::HTTP;
http_get "", sub { print $_[1] };
# ... do something else here


This module is an AnyEvent user, you need to make sure that you use and run a supported event loop.

This module implements a simple, stateless and non-blocking HTTP client. It supports GET, POST and other request methods, cookies and more, all on a very low level. It can follow redirects, supports proxies, and automatically limits the number of connections to the values specified in the RFC.

It should generally be a ``good client'' that is enough for most HTTP tasks. Simple tasks should be simple, but complex tasks should still be possible as the user retains control over request and response headers.

The caller is responsible for authentication management, cookies (if the simplistic implementation in this module doesn't suffice), referer and other high-level protocol details for which this module offers only limited support.


http_get $url, key => value..., $cb->($data, $headers)
Executes an HTTP-GET request. See the http_request function for details on additional parameters and the return value.
http_head $url, key => value..., $cb->($data, $headers)
Executes an HTTP-HEAD request. See the http_request function for details on additional parameters and the return value.
http_post $url, $body, key => value..., $cb->($data, $headers)
Executes an HTTP-POST request with a request body of $body. See the http_request function for details on additional parameters and the return value.
http_request $method => $url, key => value..., $cb->($data, $headers)
Executes a HTTP request of type $method (e.g. "GET", "POST"). The URL must be an absolute http or https URL.

When called in void context, nothing is returned. In other contexts, "http_request" returns a ``cancellation guard'' - you have to keep the object at least alive until the callback get called. If the object gets destroyed before the callback is called, the request will be cancelled.

The callback will be called with the response body data as first argument (or "undef" if an error occurred), and a hash-ref with response headers (and trailers) as second argument.

All the headers in that hash are lowercased. In addition to the response headers, the ``pseudo-headers'' (uppercase to avoid clashing with possible response headers) "HTTPVersion", "Status" and "Reason" contain the three parts of the HTTP Status-Line of the same name. If an error occurs during the body phase of a request, then the original "Status" and "Reason" values from the header are available as "OrigStatus" and "OrigReason".

The pseudo-header "URL" contains the actual URL (which can differ from the requested URL when following redirects - for example, you might get an error that your URL scheme is not supported even though your URL is a valid http URL because it redirected to an ftp URL, in which case you can look at the URL pseudo header).

The pseudo-header "Redirect" only exists when the request was a result of an internal redirect. In that case it is an array reference with the "($data, $headers)" from the redirect response. Note that this response could in turn be the result of a redirect itself, and "$headers->{Redirect}[1]{Redirect}" will then contain the original response, and so on.

If the server sends a header multiple times, then their contents will be joined together with a comma (","), as per the HTTP spec.

If an internal error occurs, such as not being able to resolve a hostname, then $data will be "undef", "$headers->{Status}" will be 590-599 and the "Reason" pseudo-header will contain an error message. Currently the following status codes are used:

595 - errors during connection establishment, proxy handshake.
596 - errors during TLS negotiation, request sending and header processing.
597 - errors during body receiving or processing.
598 - user aborted request via "on_header" or "on_body".
599 - other, usually nonretryable, errors (garbled URL etc.).

A typical callback might look like this:

   sub {
      my ($body, $hdr) = @_;
      if ($hdr->{Status} =~ /^2/) {
         ... everything should be ok
      } else {
         print "error, $hdr->{Status} $hdr->{Reason}\n";

Additional parameters are key-value pairs, and are fully optional. They include:

recurse => $count (default: $MAX_RECURSE)
Whether to recurse requests or not, e.g. on redirects, authentication and other retries and so on, and how often to do so.

Only redirects to http and https URLs are supported. While most common redirection forms are handled entirely within this module, some require the use of the optional URI module. If it is required but missing, then the request will fail with an error.

headers => hashref
The request headers to use. Currently, "http_request" may provide its own "Host:", "Content-Length:", "Connection:" and "Cookie:" headers and will provide defaults at least for "TE:", "Referer:" and "User-Agent:" (this can be suppressed by using "undef" for these headers in which case they won't be sent at all).

You really should provide your own "User-Agent:" header value that is appropriate for your program - I wouldn't be surprised if the default AnyEvent string gets blocked by webservers sooner or later.

Also, make sure that your headers names and values do not contain any embedded newlines.

timeout => $seconds
The time-out to use for various stages - each connect attempt will reset the timeout, as will read or write activity, i.e. this is not an overall timeout.

Default timeout is 5 minutes.

proxy => [$host, $port[, $scheme]] or undef
Use the given http proxy for all requests, or no proxy if "undef" is used.

$scheme must be either missing or must be "http" for HTTP.

If not specified, then the default proxy is used (see "AnyEvent::HTTP::set_proxy").

Currently, if your proxy requires authorization, you have to specify an appropriate ``Proxy-Authorization'' header in every request.

body => $string
The request body, usually empty. Will be sent as-is (future versions of this module might offer more options).
cookie_jar => $hash_ref
Passing this parameter enables (simplified) cookie-processing, loosely based on the original netscape specification.

The $hash_ref must be an (initially empty) hash reference which will get updated automatically. It is possible to save the cookie jar to persistent storage with something like JSON or Storable - see the "AnyEvent::HTTP::cookie_jar_expire" function if you wish to remove expired or session-only cookies, and also for documentation on the format of the cookie jar.

Note that this cookie implementation is not meant to be complete. If you want complete cookie management you have to do that on your own. "cookie_jar" is meant as a quick fix to get most cookie-using sites working. Cookies are a privacy disaster, do not use them unless required to.

When cookie processing is enabled, the "Cookie:" and "Set-Cookie:" headers will be set and handled by this module, otherwise they will be left untouched.

tls_ctx => $scheme | $tls_ctx
Specifies the AnyEvent::TLS context to be used for https connections. This parameter follows the same rules as the "tls_ctx" parameter to AnyEvent::Handle, but additionally, the two strings "low" or "high" can be specified, which give you a predefined low-security (no verification, highest compatibility) and high-security (CA and common-name verification) TLS context.

The default for this option is "low", which could be interpreted as ``give me the page, no matter what''.

See also the "sessionid" parameter.

session => $string
The module might reuse connections to the same host internally. Sometimes (e.g. when using TLS), you do not want to reuse connections from other sessions. This can be achieved by setting this parameter to some unique ID (such as the address of an object storing your state data, or the TLS context) - only connections using the same unique ID will be reused.
on_prepare => $callback->($fh)
In rare cases you need to ``tune'' the socket before it is used to connect (for example, to bind it on a given IP address). This parameter overrides the prepare callback passed to "AnyEvent::Socket::tcp_connect" and behaves exactly the same way (e.g. it has to provide a timeout). See the description for the $prepare_cb argument of "AnyEvent::Socket::tcp_connect" for details.
tcp_connect => $callback->($host, $service, $connect_cb, $prepare_cb)
In even rarer cases you want total control over how AnyEvent::HTTP establishes connections. Normally it uses AnyEvent::Socket::tcp_connect to do this, but you can provide your own "tcp_connect" function - obviously, it has to follow the same calling conventions, except that it may always return a connection guard object.

There are probably lots of weird uses for this function, starting from tracing the hosts "http_request" actually tries to connect, to (inexact but fast) host => IP address caching or even socks protocol support.

on_header => $callback->($headers)
When specified, this callback will be called with the header hash as soon as headers have been successfully received from the remote server (not on locally-generated errors).

It has to return either true (in which case AnyEvent::HTTP will continue), or false, in which case AnyEvent::HTTP will cancel the download (and call the finish callback with an error code of 598).

This callback is useful, among other things, to quickly reject unwanted content, which, if it is supposed to be rare, can be faster than first doing a "HEAD" request.

The downside is that cancelling the request makes it impossible to re-use the connection. Also, the "on_header" callback will not receive any trailer (headers sent after the response body).

Example: cancel the request unless the content-type is ``text/html''.

   on_header => sub {
      $_[0]{"content-type"} =~ /^text\/html\s*(?:;|$)/
on_body => $callback->($partial_body, $headers)
When specified, all body data will be passed to this callback instead of to the completion callback. The completion callback will get the empty string instead of the body data.

It has to return either true (in which case AnyEvent::HTTP will continue), or false, in which case AnyEvent::HTTP will cancel the download (and call the completion callback with an error code of 598).

The downside to cancelling the request is that it makes it impossible to re-use the connection.

This callback is useful when the data is too large to be held in memory (so the callback writes it to a file) or when only some information should be extracted, or when the body should be processed incrementally.

It is usually preferred over doing your own body handling via "want_body_handle", but in case of streaming APIs, where HTTP is only used to create a connection, "want_body_handle" is the better alternative, as it allows you to install your own event handler, reducing resource usage.

want_body_handle => $enable
When enabled (default is disabled), the behaviour of AnyEvent::HTTP changes considerably: after parsing the headers, and instead of downloading the body (if any), the completion callback will be called. Instead of the $body argument containing the body data, the callback will receive the AnyEvent::Handle object associated with the connection. In error cases, "undef" will be passed. When there is no body (e.g. status 304), the empty string will be passed.

The handle object might or might not be in TLS mode, might be connected to a proxy, be a persistent connection, use chunked transfer encoding etc., and configured in unspecified ways. The user is responsible for this handle (it will not be used by this module anymore).

This is useful with some push-type services, where, after the initial headers, an interactive protocol is used (typical example would be the push-style twitter API which starts a JSON/XML stream).

If you think you need this, first have a look at "on_body", to see if that doesn't solve your problem in a better way.

persistent => $boolean
Try to create/reuse a persistent connection. When this flag is set (default: true for idempotent requests, false for all others), then "http_request" tries to re-use an existing (previously-created) persistent connection to the host and, failing that, tries to create a new one.

Requests failing in certain ways will be automatically retried once, which is dangerous for non-idempotent requests, which is why it defaults to off for them. The reason for this is because the bozos who designed HTTP/1.1 made it impossible to distinguish between a fatal error and a normal connection timeout, so you never know whether there was a problem with your request or not.

When reusing an existent connection, many parameters (such as TLS context) will be ignored. See the "session" parameter for a workaround.

keepalive => $boolean
Only used when "persistent" is also true. This parameter decides whether "http_request" tries to handshake a HTTP/1.0-style keep-alive connection (as opposed to only a HTTP/1.1 persistent connection).

The default is true, except when using a proxy, in which case it defaults to false, as HTTP/1.0 proxies cannot support this in a meaningful way.

handle_params => { key => value ... }
The key-value pairs in this hash will be passed to any AnyEvent::Handle constructor that is called - not all requests will create a handle, and sometimes more than one is created, so this parameter is only good for setting hints.

Example: set the maximum read size to 4096, to potentially conserve memory at the cost of speed.

   handle_params => {
      max_read_size => 4096,

Example: do a simple HTTP GET request for and print the response body.

   http_request GET => "", sub {
      my ($body, $hdr) = @_;
      print "$body\n";

Example: do a HTTP HEAD request on, use a timeout of 30 seconds.

      HEAD    => "",
      headers => { "user-agent" => "MySearchClient 1.0" },
      timeout => 30,
      sub {
         my ($body, $hdr) = @_;
         use Data::Dumper;
         print Dumper $hdr;

Example: do another simple HTTP GET request, but immediately try to cancel it.

   my $request = http_request GET => "", sub {
      my ($body, $hdr) = @_;
      print "$body\n";
   undef $request;


AnyEvent::HTTP uses the AnyEvent::Socket::tcp_connect function for the actual connection, which in turn uses AnyEvent::DNS to resolve hostnames. The latter is a simple stub resolver and does no caching on its own. If you want DNS caching, you currently have to provide your own default resolver (by storing a suitable resolver object in $AnyEvent::DNS::RESOLVER) or your own "tcp_connect" callback.


AnyEvent::HTTP::set_proxy "proxy-url"
Sets the default proxy server to use. The proxy-url must begin with a string of the form "http://host:port", croaks otherwise.

To clear an already-set proxy, use "undef".

When AnyEvent::HTTP is loaded for the first time it will query the default proxy from the operating system, currently by looking at "$ENV{http_proxy"}.

AnyEvent::HTTP::cookie_jar_expire $jar[, $session_end]
Remove all cookies from the cookie jar that have been expired. If $session_end is given and true, then additionally remove all session cookies.

You should call this function (with a true $session_end) before you save cookies to disk, and you should call this function after loading them again. If you have a long-running program you can additionally call this function from time to time.

A cookie jar is initially an empty hash-reference that is managed by this module. Its format is subject to change, but currently it is as follows:

The key "version" has to contain 1, otherwise the hash gets emptied. All other keys are hostnames or IP addresses pointing to hash-references. The key for these inner hash references is the server path for which this cookie is meant, and the values are again hash-references. Each key of those hash-references is a cookie name, and the value, you guessed it, is another hash-reference, this time with the key-value pairs from the cookie, except for "expires" and "max-age", which have been replaced by a "_expires" key that contains the cookie expiry timestamp. Session cookies are indicated by not having an "_expires" key.

Here is an example of a cookie jar with a single cookie, so you have a chance of understanding the above paragraph:

      version    => 1,
      "" => {
         "/" => {
            "mythweb_id" => {
              _expires => 1293917923,
              value    => "ooRung9dThee3ooyXooM1Ohm",
$date = AnyEvent::HTTP::format_date $timestamp
Takes a POSIX timestamp (seconds since the epoch) and formats it as a HTTP Date (RFC 2616).
$timestamp = AnyEvent::HTTP::parse_date $date
Takes a HTTP Date (RFC 2616) or a Cookie date (netscape cookie spec) or a bunch of minor variations of those, and returns the corresponding POSIX timestamp, or "undef" if the date cannot be parsed.
The default value for the "recurse" request parameter (default: 10).
The default timeout for connection operations (default: 300).
The default value for the "User-Agent" header (the default is "Mozilla/5.0 (compatible; U; AnyEvent-HTTP/$VERSION; +").
The maximum number of concurrent connections to the same host (identified by the hostname). If the limit is exceeded, then additional requests are queued until previous connections are closed. Both persistent and non-persistent connections are counted in this limit.

The default value for this is 4, and it is highly advisable to not increase it much.

For comparison: the RFC's recommend 4 non-persistent or 2 persistent connections, older browsers used 2, newer ones (such as firefox 3) typically use 6, and Opera uses 8 because like, they have the fastest browser and give a shit for everybody else on the planet.

The time after which idle persistent connections get closed by AnyEvent::HTTP (default: 3).
The number of active connections. This is not the number of currently running requests, but the number of currently open and non-idle TCP connections. This number can be useful for load-leveling.


This section contains some more elaborate ``real-world'' examples or code snippets.


Downloading files with HTTP can be quite tricky, especially when something goes wrong and you want to resume.

Here is a function that initiates and resumes a download. It uses the last modified time to check for file content changes, and works with many HTTP/1.0 servers as well, and usually falls back to a complete re-download on older servers.

It calls the completion callback with either "undef", which means a nonretryable error occurred, 0 when the download was partial and should be retried, and 1 if it was successful.

   use AnyEvent::HTTP;
   sub download($$$) {
      my ($url, $file, $cb) = @_;
      open my $fh, "+<", $file
         or die "$file: $!";
      my %hdr;
      my $ofs = 0;
      warn stat $fh;
      warn -s _;
      if (stat $fh and -s _) {
         $ofs = -s _;
         warn "-s is ", $ofs;
         $hdr{"if-unmodified-since"} = AnyEvent::HTTP::format_date +(stat _)[9];
         $hdr{"range"} = "bytes=$ofs-";
      http_get $url,
         headers   => \%hdr,
         on_header => sub {
            my ($hdr) = @_;
            if ($hdr->{Status} == 200 && $ofs) {
               # resume failed
               truncate $fh, $ofs = 0;
            sysseek $fh, $ofs, 0;
         on_body   => sub {
            my ($data, $hdr) = @_;
            if ($hdr->{Status} =~ /^2/) {
               length $data == syswrite $fh, $data
                  or return; # abort on write errors
         sub {
            my (undef, $hdr) = @_;
            my $status = $hdr->{Status};
            if (my $time = AnyEvent::HTTP::parse_date $hdr->{"last-modified"}) {
               utime $fh, $time, $time;
            if ($status == 200 || $status == 206 || $status == 416) {
               # download ok || resume ok || file already fully downloaded
               $cb->(1, $hdr);
            } elsif ($status == 412) {
               # file has changed while resuming, delete and retry
               unlink $file;
               $cb->(0, $hdr);
            } elsif ($status == 500 or $status == 503 or $status =~ /^59/) {
               # retry later
               $cb->(0, $hdr);
            } else {
               $cb->(undef, $hdr);
   download "http://server/somelargefile", "/tmp/somelargefile", sub {
      if ($_[0]) {
         print "OK!\n";
      } elsif (defined $_[0]) {
         print "please retry later\n";
      } else {
         print "ERROR\n";


Socks proxies are not directly supported by AnyEvent::HTTP. You can compile your perl to support socks, or use an external program such as socksify (dante) or tsocks to make your program use a socks proxy transparently.

Alternatively, for AnyEvent::HTTP only, you can use your own "tcp_connect" function that does the proxy handshake - here is an example that works with socks4a proxies:

   use Errno;
   use AnyEvent::Util;
   use AnyEvent::Socket;
   use AnyEvent::Handle;
   # host, port and username of/for your socks4a proxy
   my $socks_host = "";
   my $socks_port = 9050;
   my $socks_user = "";
   sub socks4a_connect {
      my ($host, $port, $connect_cb, $prepare_cb) = @_;
      my $hdl = new AnyEvent::Handle
         connect    => [$socks_host, $socks_port],
         on_prepare => sub { $prepare_cb->($_[0]{fh}) },
         on_error   => sub { $connect_cb->() },
      $hdl->push_write (pack "CCnNZ*Z*", 4, 1, $port, 1, $socks_user, $host);
      $hdl->push_read (chunk => 8, sub {
         my ($hdl, $chunk) = @_;
         my ($status, $port, $ipn) = unpack "xCna4", $chunk;
         if ($status == 0x5a) {
            $connect_cb->($hdl->{fh}, (format_address $ipn) . ":$port");
         } else {
            $! = Errno::ENXIO; $connect_cb->();

Use "socks4a_connect" instead of "tcp_connect" when doing "http_request"s, possibly after switching off other proxy types:

   AnyEvent::HTTP::set_proxy undef; # usually you do not want other proxies
   http_get '', tcp_connect => \&socks4a_connect, sub {
      my ($data, $headers) = @_;


   Marc Lehmann <[email protected]>

With many thanks to XXXXXXX XXXXXXX, who provided countless testcases and bugreports.