Provided by: libhijk-perl_0.27-1_all 
NAME
Hijk - Fast & minimal low-level HTTP client
SYNOPSIS
A simple GET request:
use Hijk ();
my $res = Hijk::request({
method => "GET",
host => "example.com",
port => "80",
path => "/flower",
query_string => "color=red"
});
if (exists $res->{error} and $res->{error} & Hijk::Error::TIMEOUT) {
die "Oh noes we had some sort of timeout";
}
die "Expecting an 'OK' response" unless $res->{status} == 200;
say $res->{body};
A POST request, you have to manually set the appropriate headers, URI escape your values etc.
use Hijk ();
use URI::Escape qw(uri_escape);
my $res = Hijk::request({
method => "POST",
host => "example.com",
port => "80",
path => "/new",
head => [ "Content-Type" => "application/x-www-form-urlencoded" ],
query_string => "type=flower&bucket=the%20one%20out%20back",
body => "description=" . uri_escape("Another flower, let's hope it's exciting"),
});
die "Expecting an 'OK' response" unless $res->{status} == 200;
DESCRIPTION
Hijk is a fast & minimal low-level HTTP client intended to be used where you control both the client and
the server, e.g. for talking to some internal service from a frontend user-facing web application.
It is "NOT" a general HTTP user agent, it doesn't support redirects, proxies, SSL and any number of other
advanced HTTP features like (in roughly descending order of feature completeness) LWP::UserAgent,
WWW::Curl, HTTP::Tiny, HTTP::Lite or Furl. This library is basically one step above manually talking HTTP
over sockets.
Having said that it's lightning fast and extensively used in production at Booking.com
<https://www.booking.com> where it's used as the go-to transport layer for talking to internal services.
It uses non-blocking sockets and correctly handles all combinations of connect/read timeouts and other
issues you might encounter from various combinations of parts of your system going down or becoming
otherwise unavailable.
FUNCTION: Hijk::request( $args :HashRef ) :HashRef
"Hijk::request" is the only function you should use. It (or anything else in this package for that
matter) is not exported, so you have to use the fully qualified name.
It takes a "HashRef" of arguments and either dies or returns a "HashRef" as a response.
The "HashRef" argument to it must contain some of the key-value pairs from the following list. The value
for "host" and "port" are mandatory, but others are optional with default values listed below.
protocol => "HTTP/1.1", # (or "HTTP/1.0")
host => ...,
port => ...,
connect_timeout => undef,
read_timeout => undef,
read_length => 10240,
method => "GET",
path => "/",
query_string => "",
head => [],
body => "",
socket_cache => \%Hijk::SOCKET_CACHE, # (undef to disable, or \my %your_socket_cache)
on_connect => undef, # (or sub { ... })
parse_chunked => 0,
head_as_array => 0,
no_default_host_header => 1,
Notice how Hijk does not take a full URI string as input, you have to specify the individual parts of the
URL. Users who need to parse an existing URI string to produce a request should use the URI module to do
so.
The value of "head" is an "ArrayRef" of key-value pairs instead of a "HashRef", this way you can decide
in which order the headers are sent, and you can send the same header name multiple times. For example:
head => [
"Content-Type" => "application/json",
"X-Requested-With" => "Hijk",
]
Will produce these request headers:
Content-Type: application/json
X-Requested-With: Hijk
In addition Hijk will provide a "Host" header for you by default with the "host" value you pass to
"request()". To suppress this (e.g. to send custom "Host" requests) pass a true value to the
"no_default_host_header" option and provide your own "Host" header in the "head" "ArrayRef" (or don't, if
you want to construct a "Host"-less request knock yourself out...).
Hijk doesn't escape any values for you, it just passes them through as-is. You can easily produce invalid
requests if e.g. any of these strings contain a newline, or aren't otherwise properly escaped.
The value of "connect_timeout" or "read_timeout" is in floating point seconds, and is used as the time
limit for connecting to the host, and reading the response back from it, respectively. The default value
for both is "undef", meaning no timeout limit. If you don't supply these timeouts and the host really is
unreachable or slow, we'll reach the TCP timeout limit before returning some other error to you.
The default "protocol" is "HTTP/1.1", but you can also specify "HTTP/1.0". The advantage of using
"HTTP/1.1" is support for keep-alive, which matters a lot in environments where the connection setup
represents non-trivial overhead. Sometimes that overhead is negligible (e.g. on Linux talking to an nginx
on the local network), and keeping open connections down and reducing complexity is more important, in
those cases you can either use "HTTP/1.0", or specify "Connection: close" in the request, but just using
"HTTP/1.0" is an easy way to accomplish the same thing.
By default we will provide a "socket_cache" for you which is a global singleton that we maintain keyed on
"join($;, $$, $host, $port)". Alternatively you can pass in "socket_cache" hash of your own which we'll
use as the cache. To completely disable the cache pass in "undef".
The optional "on_connect" callback is intended to be used for you to figure out from production traffic
what you should set the "connect_timeout". I.e. you can start a timer when you call "Hijk::request()"
that you end when "on_connect" is called, that's how long it took us to get a connection. If you start
another timer in that callback that you end when "Hijk::request()" returns to you that'll give you how
long it took to send/receive data after we constructed the socket, i.e. it'll help you to tweak your
"read_timeout". The "on_connect" callback is provided with no arguments, and is called in void context.
We have experimental support for parsing chunked responses encoding. historically Hijk didn't support
this at all and if you wanted to use it with e.g. nginx you had to add "chunked_transfer_encoding off" to
the nginx config file.
Since you may just want to do that instead of having Hijk do more work to parse this out with a more
complex and experimental codepath you have to explicitly enable it with "parse_chunked". Otherwise Hijk
will die when it encounters chunked responses. The "parse_chunked" option may be turned on by default in
the future.
The return value is a "HashRef" representing a response. It contains the following key-value pairs.
proto => :Str
status => :StatusCode
body => :Str
head => :HashRef (or :ArrayRef with "head_as_array")
error => :PositiveInt
error_message => :Str
errno_number => :Int
errno_string => :Str
For example, to send a request to "http://example.com/flower?color=red", pass the following parameters:
my $res = Hijk::request({
host => "example.com",
port => "80",
path => "/flower",
query_string => "color=red"
});
die "Response is not 'OK'" unless $res->{status} == 200;
Notice that you do not need to put the leading "?" character in the "query_string". You do, however, need
to properly "uri_escape" the content of "query_string".
Again, Hijk doesn't escape any values for you, so these values MUST be properly escaped before being
passed in, unless you want to issue invalid requests.
By default the "head" in the response is a "HashRef" rather then an "ArrayRef". This makes it easier to
retrieve specific header fields, but it means that we'll clobber any duplicated header names with the
most recently seen header value. To get the returned headers as an "ArrayRef" instead specify
"head_as_array".
If you want to fiddle with the "read_length" value it controls how much we "POSIX::read($fd, $buf,
$read_length)" at a time.
We currently don't support servers returning a http body without an accompanying "Content-Length" header;
bodies MUST have a "Content-Length" or we won't pick them up.
ERROR CODES
If we had a recoverable error we'll include an "error" key whose value is a bitfield that you can check
against Hijk::Error::* constants. Those are:
Hijk::Error::CONNECT_TIMEOUT
Hijk::Error::READ_TIMEOUT
Hijk::Error::TIMEOUT
Hijk::Error::CANNOT_RESOLVE
Hijk::Error::REQUEST_SELECT_ERROR
Hijk::Error::REQUEST_WRITE_ERROR
Hijk::Error::REQUEST_ERROR
Hijk::Error::RESPONSE_READ_ERROR
Hijk::Error::RESPONSE_BAD_READ_VALUE
Hijk::Error::RESPONSE_ERROR
In addition we might return "error_message", "errno_number" and "errno_string" keys, see the discussion
of "Hijk::Error::REQUEST_*" and "Hijk::Error::RESPONSE_*" errors below.
The "Hijk::Error::TIMEOUT" constant is the same as "Hijk::Error::CONNECT_TIMEOUT |
Hijk::Error::READ_TIMEOUT". It's there for convenience so you can do:
.. if exists $res->{error} and $res->{error} & Hijk::Error::TIMEOUT;
Instead of the more verbose:
.. if exists $res->{error} and $res->{error} & (Hijk::Error::CONNECT_TIMEOUT | Hijk::Error::READ_TIMEOUT)
We'll return "Hijk::Error::CANNOT_RESOLVE" if we can't "gethostbyname()" the host you've provided.
If we fail to do a "select()" or "write()" during when sending the response we'll return
"Hijk::Error::REQUEST_SELECT_ERROR" or "Hijk::Error::REQUEST_WRITE_ERROR", respectively. Similarly to
"Hijk::Error::TIMEOUT" the "Hijk::Error::REQUEST_ERROR" constant is a union of these two, and any other
request errors we might add in the future.
When we're getting the response back we'll return "Hijk::Error::RESPONSE_READ_ERROR" when we can't
"read()" the response, and "Hijk::Error::RESPONSE_BAD_READ_VALUE" when the value we got from "read()" is
0. The "Hijk::Error::RESPONSE_ERROR" constant is a union of these two and any other response errors we
might add in the future.
Some of these "Hijk::Error::REQUEST_*" and "Hijk::Error::RESPONSE_*" errors are re-thrown errors from
system calls. In that case we'll also pass along "error_message" which is a short human readable error
message about the error, as well as "errno_number" & "errno_string", which are "$!+0" and "$!" at the
time we had the error.
Hijk might encounter other errors during the course of the request and WILL call "die" if that happens,
so if you don't want your program to stop when a request like that fails wrap it in "eval".
Having said that the point of the "Hijk::Error::*" interface is that all errors that happen during normal
operation, i.e. making valid requests against servers where you can have issues like timeouts, network
blips or the server thread on the other end being suddenly kill -9'd should be caught, categorized and
returned in a structural way by Hijk.
We're not currently aware of any issues that occur in such normal operations that aren't classified as a
"Hijk::Error::*", and if we find new issues that fit the criteria above we'll likely just make a new
"Hijk::Error::*" for it.
We're just not trying to guarantee that the library can never "die", and aren't trying to catch truly
exceptional issues like e.g. "fcntl()" failing on a valid socket.
AUTHORS
Kang-min Liu <gugod@gugod.org>
Ævar Arnfjörð Bjarmason <avar@cpan.org>
Borislav Nikolov <jack@sofialondonmoskva.com>
Damian Gryski <damian@gryski.com>
COPYRIGHT
Copyright (c) 2013 Kang-min Liu "<gugod@gugod.org>".
LICENCE
The MIT License
DISCLAIMER OF WARRANTY
BECAUSE THIS SOFTWARE IS LICENSED FREE OF CHARGE, THERE IS NO WARRANTY FOR THE SOFTWARE, TO THE EXTENT
PERMITTED BY APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR OTHER
PARTIES PROVIDE THE SOFTWARE "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED,
INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE SOFTWARE IS WITH YOU. SHOULD THE
SOFTWARE PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING, REPAIR, OR CORRECTION.
IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING WILL ANY COPYRIGHT HOLDER, OR ANY
OTHER PARTY WHO MAY MODIFY AND/OR REDISTRIBUTE THE SOFTWARE AS PERMITTED BY THE ABOVE LICENCE, BE LIABLE
TO YOU FOR DAMAGES, INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES ARISING OUT OF
THE USE OR INABILITY TO USE THE SOFTWARE (INCLUDING BUT NOT LIMITED TO LOSS OF DATA OR DATA BEING
RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD PARTIES OR A FAILURE OF THE SOFTWARE TO OPERATE
WITH ANY OTHER SOFTWARE), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
DAMAGES.