NAME¶
"IO::Async::Sequencer" - handle a serial pipeline of requests /
responses (EXPERIMENTAL)
SYNOPSIS¶
When used as a client:
use IO::Async::Loop;
my $loop = IO::Async::Loop->new;
my $sock = ...
my $sequencer = IO::Async::Sequencer->new(
handle => $sock,
on_read => sub {
my ( $self, $buffref, $eof ) = @_;
return 0 unless $$buffref =~ s/^(.*)\n//;
my $line = $1;
$line =~ m/^RESPONSE (.*)$/ and
$self->incoming_response( $1 ), return 1;
print STDERR "An error - didn't recognise the line $line\n";
},
marshall_request => sub {
my ( $self, $request ) = @_;
return "REQUEST $request\n";
},
);
$loop->add( $sequencer );
$sequencer->request(
request => "hello",
on_response => sub {
my ( $response ) = @_;
print "The response is: $response\n";
},
);
When used as a server:
my $sequencer = IO::Async::Sequencer->new(
handle => $sock,
on_read => sub {
my ( $self, $buffref, $eof ) = @_;
return 0 unless $$buffref =~ s/^(.*)\n//;
my $line = $1;
$line =~ m/^REQUEST (.*)$/ and
$self->incoming_request( $1 ), return 1;
print STDERR "An error - didn't recognise the line $line\n";
},
on_request => sub {
my ( $self, $token, $request ) = @_;
# Now to invoke the application logic, whatever it may be
solve_request(
request => $request,
on_completed => sub {
my ( $response ) = @_;
$self->respond( $token, $response );
}
);
},
marshall_response => sub {
my ( $self, $response ) = @_;
return "RESPONSE $response\n";
},
);
DESCRIPTION¶
This module provides an EXPERIMENTAL subclass of "IO::Async::Stream"
which may be helpful in implementing serial pipeline-based network protocols
of requests and responses. It deals with low-level details such as pairing up
responses to requests in ordered protocols, and allows a convenient location
to store the line stream marshalling and demarshalling code.
The intention is that individual network protocols may be implemented as a
subclass of this class, providing the marshalling and demarshalling code
itself, providing a clean implementation to be used by the using code. An
example protocol that would be easy to implement in this way would be HTTP.
Objects in this class can operate in any of three ways:
- •
- A pure client
The object is asked to send requests by the containing code and invokes
response handling code when responses arrive.
- •
- A pure server
The object receives requests from the filehandle, processes them, and sends
responses back, but does not initiate any traffic of its own.
- •
- Mixed
The object behaves as a mixture of the two; initiating requests, as well as
responding to those of its peer connection.
The exact mode of operation of any object is not declared explicitly, but
instead is an artefact of the set of callbacks provided to the constructor or
methods defined by the subclass. Certain callbacks or methods only make sense
for one mode or the other.
The various operations required can each be provided as callback functions given
in keys to the constructor, or as object methods on a subclass of this class.
Keys passed to the constructor will take precidence over defined methods.
As it is still EXPERIMENTAL, any details of this class are liable to change in
future versions. It shouldn't yet be relied upon as a stable interface.
PARAMETERS¶
The following named parameters may be passed to "new" or
"configure":
- on_read => CODE
- As for IO::Async::Stream. The code here should invoke the
"incoming_request" or "incoming_response" methods when
appropriate, after having parsed the incoming stream. See the SYNOPSIS or
EXAMPLES sections for more detail.
Each request can optionally provide its own handler for reading its
response, using the "on_read" key to the "request"
method. The handler provided to the constructor is only used if this is
not provided.
- on_request => CODE
- A callback that is invoked when the
"incoming_request" method is called (i.e. when operating in
server mode). It is passed the request, and a token to identify it when
sending a response.
$on_request->( $self, $token, $request );
The token should be considered as an opaque value - passed into the
"respond" method when a response is ready, but not otherwise
used or modified.
- marshall_request => CODE
- marshall_response => CODE
- Callbacks that the "request" or
"respond" methods will use, respectively, to stream a request or
response object into a string of bytes to write to the underlying file
handle.
$string = $marshall_request->( $self, $request );
$string = $marshall_response->( $self, $response );
These are used respectively by the client and server modes.
- pipeline => BOOL
- Optional. Controls whether requests will be pipelined; that
is, all requests will be sent by the client before responses are received.
If this option is disabled, only the first request will be sent. Other
requests will be queued internally, and each will be sent when the
response to the previous has been received. Defaults enabled; supply a
defined but false value to disable.
SUBCLASS METHODS¶
This class is intended as a base class for building specific protocol handling
code on top of. These methods are intended to be called by the specific
subclass, rather than the containing application code.
$sequencer->incoming_request( $request )¶
To be called from "on_read".
This method informs the sequencer that a new request has arrived. It will invoke
"on_request", passing in a token to identify the request for stream
ordering purposes, and the request itself.
$sequencer->incoming_response( $response )¶
To be called from "on_read".
This method informs the sequencer that a response has arrived. It will invoke
"on_response" that had been passed to the "request" method
that sent the original request.
PUBLIC METHODS¶
These methods are intended to be called by the application code using a subclass
of this class.
$sequencer->request( %params )¶
Called in client mode, this method sends a request upstream, and awaits a
response to it. Can be called in one of two ways; either giving a specific
"on_read" handler to be used when the response to this request is
expected, or by providing an "on_response" handler for when the
default handler invokes "incoming_response".
The %params hash takes the following arguments:
- request => SCALAR
- The request value to pass to
"marshall_request".
- on_response => CODE
- A continuation to invoke when a response to this request
arrives from the upstream server. It will be invoked as
$on_response->( $response );
- on_read => CODE
- A callback to use to parse the incoming stream while the
response to this particular request is expected. It will be invoked the
same as for "IO::Async::Stream"; i.e.
$on_read->( $self, $buffref, $eof )
This handler should return "undef" when it has finished handling
the response, so that the next one queued can be invoked (or the default
if none exists). It MUST NOT call "incoming_response". Instead,
the code should directly implement the behaviour for receipt of a
response.
If the "on_read" key is used, it is intended that a specific subclass
that implements a specific protocol would construct the callback code in a
method it provides, intended for the using code to call.
$sequencer->respond( $token, $response )¶
Called in server mode, usually at the end of "on_request", or some
continuation created within it, this method sends a response back downstream
to a client that had earlier requested it.
- $token
- The token that was passed into the "on_request".
Used to ensure responses are sent in the right order.
- $response
- The response value to pass to
"marshall_response".
EXAMPLES¶
A simple line-based server¶
The following sequencer implements a simple server which takes and responds with
CRLF-delimited lines.
package LineSequencer;
use base qw( IO::Async::Sequencer );
my $CRLF = "\x0d\x0a"; # More portable than \r\n
sub on_read {
my ( $self, $buffref, $eof ) = @_;
while( $buffref =~ s/^(.*)$CRLF// ) {
$self->incoming_request( $1 );
}
return 0;
}
sub marshall_response {
my ( $self, $response ) = @_;
return $response . $CRLF;
}
1;
The server could then be used, for example, as a simple echo server that replies
whatever the client said, in uppercase. This would be done using an
"on_request" like the following.
my $linesequencer = LineSequencer->new(
handle => ...
on_request => sub {
my ( $self, $token, $request ) = @_;
$self->respond( $token, uc $request );
}
);
It is likely, however, that any real use of the server in a non-trivial way
would perform much more work than this, and only call
"$self->respond" in an eventual continuation at the end of
performing its work. The $token is used to identify the request that the
response responds to, so that it can be sent in the correct order.
Per-request "on_read" handler¶
If an "on_read" handler is provided to the "request" method
in client mode, then that handler will be used when the response to that
request is expected to arrive. This will be used instead of the
"incoming_response" method and the "on_response" handler.
If every request provides its own handler, then the one in the constructor
would only be used for unrequested input from the server - perhaps to generate
an error condition of some kind.
my $sequencer = IO::Async::Sequencer->new(
...
on_read => sub {
my ( $self, $buffref, $eof ) = @_;
print STDERR "Spurious input: $$buffref\n";
$self->close;
return 0;
},
marshall_request => sub {
my ( $self, $request ) = @_;
return "GET $request" . $CRLF;
},
);
$sequencer->request(
request => "some key",
on_read => sub {
my ( $self, $buffref, $eof ) = @_;
return 0 unless $$buffref =~ s/^(.*)$CRLF//;
my $line = $1;
print STDERR "Got response: $1\n" if $line =~ m/^HAVE (.*)$/;
return undef; # To indicate that this response is finished
}
);
TODO¶
- •
- Some consideration of streaming errors. How does the
"on_read" signal to the containing object that a stream error
has occured? Is it fatal? Can resynchronisation be attempted later?
- •
- Support, either here or in a different class, for
out-of-order protocols, such as IMAP, where responses can arrive in a
different order than the requests were sent.
AUTHOR¶
Paul Evans <leonerd@leonerd.org.uk>