Plack::App::CGIBin::Streaming - allow old style CGI applications to use the plack streaming protocol
in your app.psgi:
use Plack::App::CGIBin::Streaming;
Plack::App::CGIBin::Streaming->new(root=>...)->to_app;
With Plack already comes Plack::App::CGIBin.
Plack::App::CGIBin::Streaming serves a very similar purpose.
So, why do I need another module? The reason is that Plack::App::CGIBin first collects all the output from your CGI scripts before it prints the first byte to the client. This renders the following simple clock script useless:
use strict;
use warnings;
$|=0;
my $boundary='The Final Frontier';
print <<"EOF";
Status: 200
Content-Type: multipart/x-mixed-replace;boundary="$boundary";
EOF
$boundary="--$boundary\n";
my $mpheader=<<'HEADER';
Content-type: text/html; charset=UTF-8;
HEADER
for(1..100) {
print ($boundary, $mpheader,
'<html><body><h1>'.localtime()."</h1></body></html>\n");
$|=1; $|=0;
sleep 1;
}
print ($boundary);
Although multipart HTTP messages are quite exotic, there are situations
where you rather want to prevent this buffering. If your document is very
large for example, each instance of your plack server allocates the RAM
to buffer it. Also, you might perhaps send out the <head> section
of your HTTP document as fast as possible to enable the browser load JS and
CSS while the plack server is still busy with producing the actual document.
Plack::App::CGIBin::Streaming compiles the CGI scripts using
CGI::Compile and provides a runtime environment similar to
Plack::App::CGIBin. Compiled scripts are cached. For production
environments, it is possible to precompile and cache scripts at server
start time, see the preload option below.
Every single request is represented as an object that inherits from Plack::App::CGIBin::Streaming::Request. This class mainly provides means for handling response headers and body.
The plack app is built as usual:
$app=Plack::App::CGIBin::Streaming->new(@options)->to_app;
@options is a list of key/value pairs configuring the app. The
Plack::App::CGIBin::Streaming class inherits from Plack::App::File.
So, everything recognized by this class is accepted. In particular, the
root parameter is used to specify the directory where your CGI programs
reside.
Additionally, these parameters are accepted:
-
request_class
specifies the class of the request object to construct for every request. This class should implement the interface described in Plack::App::CGIBin::Streaming::Request. Best if your request class inherits from Plack::App::CGIBin::Streaming::Request.
This parameter is optional. By default
Plack::App::CGIBin::Streaming::Requestis used. -
request_params
specifies a list of additional parameters to be passed to the request constructor.
By default the request constructor is passed 2 parameters. This list is appended to the parameter list like:
$R = $class->new( env => $env, responder => $responder, @{$self->request_params//[]}, ); -
preload
In a production environment, you probably want to use a (pre)forking server to run the application. In this case is is sensible to compile as much perl code as possible at server startup time by the parent process because then all the children share the RAM pages where the code resides (by copy-on-write) and you utilize your server resources much better.
One way to achieve that is to keep your CGI applications very slim and put all the actual work into modules. These modules are then
used orrequired in yourapp.psgifile.As a simpler alternative you can specify a list of
globpatterns aspreloadvalue.Plack::App::CGIBin::Streamingwill then load and compile all the scripts matching all the patterns when the app object is created.This technique has benefits and drawbacks:
-
pro: more concurrent worker children in less RAM
see above
-
con: no way to reload the application on the fly
when your scripts change you have to restart the server. Without preloading anything you could just kill all the worker children (or signal them to do so after the next request).
-
pro/con: increased privileges while preloading
the HTTP standard port is 80 and, thus, requires root privileges to bind to. scripts are preloaded before the server opens the port. So, even if it later drops privilges, at preload time you still are root.
-
Additional to the environment provided by CGI::Compile, this module provides:
-
the global variable
$Plack::App::CGIBin::Streaming::RFor the request lifetime it contains the actual request object. This variable is
localized. There is also a way to access this variable as class method.If you use a Coro based plack server, make sure to replace the guts of this variable when switching threads, see
swap_sv()in Coro::State. -
Plack::App::CGIBin::Streaming->requestorPlack::App::CGIBin::Streaming::requestThis function/method returns the current request object or
undefif called outside the request loop. -
%ENVis populatedeverything from the plack environment except keys starting with
plackorpsgi.is copied to%ENV. -
STDINandSTDOUTBoth,
STDINandSTDOUTare configured to use the Plack::App::CGIBin::Streaming::IO PerlIO layer. On output, the layer captures the data and sends it to the request object. Flushing via$|is also supported. On input, the layer simply converts calls likereadline STDINinto a method call on the underlying object.You can use PerlIO layers to turn the handles into UTF8 mode. However, refrain from using a simple
binmodeto reverse the effects of a priorbinmode STDOUT, ':utf8'. This won't pop the Plack::App::CGIBin::Streaming::IO layer but neither will it turn off UTF8 mode. This is considered a bug that I don't know how to fix. (See also below)Reading from
STDINusing UTF8 mode is also supported.
During the implementation I found a wierd bug. At least on Linux, perl
supports CHLD and CLD as name of the signal that is sent when a child
process exits. Also, when Perl calls a signal handler, it passes the signal
name as the first parameter. Now the question arises, which name is passed
when a child exits. As it happens the first assignment to %SIG{CHLD}
or $SIG{CLD} determines that name for the rest of the lifetime of the
process. Now, several plack server implementations, e.g. Starman,
rely on that name to be CHLD.
As a workaround, Plack::App::CGIBin::Streaming contains this code:
BEGIN {
local $SIG{CHLD}=$SIG{CHLD};
}
If your server dies when it receives a SIGCHLD, perhaps the module is loaded too late.
Sometimes one needs to switch STDOUT into UTF8 mode and back. Especially the
back is problematic because the way it is done is often simply
binmode STDOUT. Currently, this won't revert the effect of a previous
binmode STDOUT, ':utf8'.
Instead use:
binmode STDOUT, ':bytes';
This distribution contains a complete example in the eg/ directory.
After building the module by
perl Build.PL
./Build
you can try it out:
(cd eg && starman -l :5091 --workers=2 --preload-app app.psgi) &
Then you should be able to access
The clock example is basically the script displayed above. It works in Firefox. Other browsers don't support multipart HTTP messages.
The flush example demonstrates filtering. It has been tested wich Chromium
35 on Linux. The script first prints a part of the page that contains the
HTML comment <!-- FlushHead -->. The filter recognizes this token
and pushes the page out. You should see a red background and the string
loading -- please wait. After 2 seconds the page should turn green and
the string should change to loaded.
All of this very much depends on browser behavior. The intent is not to
provide an example that works for all of them. Instead, the capabilities
of this module are shown. You can also test these links with curl
instead.
The example PSGI file also configures an access_log and an error_log.
Torsten Förtsch torsten.foertsch@gmx.net
Copyright 2014 Binary.com
This program is free software; you can redistribute it and/or modify it
under the terms of the the Artistic License (2.0). A copy of the full
license is provided by the LICENSE file in this distribution and can
be obtained at
