RPC::XML::Server - A sample server implementation based on RPC::XML

    use RPC::XML::Server;

    ...
    $srv = RPC::XML::Server->new(port => 9000);
    # Several of these, most likely:
    $srv->add_method(...);
    ...
    $srv->server_loop; # Never returns

This is a sample XML-RPC server built upon the RPC::XML data classes, and using HTTP::Daemon and HTTP::Response for the communication layer.

Use of the RPC::XML::Server is based on an object model. A server is instantiated from the class, methods (subroutines) are made public by adding them through the object interface, and then the server object is responsible for dispatching requests (and possibly for the HTTP listening, as well).

The following methods are provided by the RPC::XML::Server class. Unless otherwise explicitly noted, all methods return the invoking object reference upon success, and a non-reference error string upon failure.

See Content Compression below for details of how the server class manages gzip-based compression and expansion of messages.

new(OPTIONS)

Creates a new object of the class and returns the blessed reference. Depending on the options, the object will contain some combination of an HTTP listener, a pre-populated HTTP::Response object, a RPC::XML::Parser object, and a dispatch table with the set of default methods pre-loaded. The options that new accepts are passed as a hash of key/value pairs (not a hash reference). The accepted options are:

no_http

If passed with a CWtrue value, prevents the creation and storage of the HTTP::Daemon object. This allows for deployment of a server object in other environments. Note that if this is set, the server_loop method described below will silently attempt to use the Net::Server module.

no_default

If passed with a CWtrue value, prevents the loading of the default methods provided with the RPC::XML distribution. These may be later loaded using the add_default_methods interface described later. The methods themselves are described below (see The Default Methods Provided).

path

host

port

queue

These four are specific to the HTTP-based nature of the server. The path argument sets the additional URI path information that clients would use to contact the server. Internally, it is not used except in outgoing status and introspection reports. The host, port and queue arguments are passed to the HTTP::Daemon constructor if they are passed. They set the hostname, TCP/IP port, and socket listening queue, respectively. They may also be used if the server object tries to use Net::Server as an alternative server core.

xpl_path

If you plan to add methods to the server object by passing filenames to the CWadd_method call, this argument may be used to specify one or more additional directories to be searched when the passed-in filename is a relative path. The value for this must be an array reference. See also add_method and xpl_path, below.

timeout

Specify a value (in seconds) for the HTTP::Daemon server to use as a timeout value when reading request data from an inbound connection. The default value is 10 seconds. This value is not used except by HTTP::Daemon.

auto_methods

If specified and set to a true value, enables the automatic searching for a requested remote method that is unknown to the server object handling the request. If set to no (or not set at all), then a request for an unknown function causes the object instance to report an error. If the routine is still not found, the error is reported. Enabling this is a security risk, and should only be permitted by a server administrator with fully informed acknowledgement and consent.

auto_updates

If specified and set to a true value, enables the checking of the modification time of the file from which a method was originally loaded. If the file has changed, the method is re-loaded before execution is handed off. As with the auto-loading of methods, this represents a security risk, and should only be permitted by a server administrator with fully informed acknowledgement and consent.

parser

If this parameter is passed, the value following it is expected to be an array reference. The contents of that array are passed to the new method of the RPC::XML::Parser object that the server object caches for its use. See the RPC::XML::Parser manual page for a list of recognized parameters to the constructor.

message_file_thresh

If this key is passed, the value associated with it is assumed to be a numerical limit to the size of in-memory messages. Any out-bound request that would be larger than this when stringified is instead written to an anonynous temporary file, and spooled from there instead. This is useful for cases in which the request includes RPC::XML::base64 objects that are themselves spooled from file-handles. This test is independent of compression, so even if compression of a request would drop it below this threshhold, it will be spooled anyway. The file itself is unlinked after the file-handle is created, so once it is freed the disk space is immediately freed.

message_temp_dir

If a message is to be spooled to a temporary file, this key can define a specific directory in which to open those files. If this is not given, then the CWtmpdir method from the File::Spec package is used, instead. Any other keys in the options hash not explicitly used by the constructor are copied over verbatim onto the object, for the benefit of sub-classing this class. All internal keys are prefixed with CW__ to avoid confusion. Feel free to use this prefix only if you wish to re-introduce confusion.

version

Returns the version string associated with this package.

product_tokens

This returns the identifying string for the server, in the format CWNAME/VERSION consistent with other applications such as Apache and LWP. It is provided here as part of the compatibility with HTTP::Daemon that is required for effective integration with Net::Server.

url

This returns the HTTP URL that the server will be responding to, when it is in the connection-accept loop. If the server object was created without a built-in HTTP listener, then this method returns CWundef.

requests

Returns the number of requests this server object has marshalled. Note that in multi-process environments (such as Apache or Net::Server::PreFork) the value returned will only reflect the messages dispatched by the specific process itself.

response

Each instance of this class (and any subclasses that do not completely override the CWnew method) creates and stores an instance of HTTP::Response, which is then used by the HTTP::Daemon or Net::Server processing loops in constructing the response to clients. The response object has all common headers pre-set for efficiency. This method returns a reference to that object.

started([BOOL])

Gets and possibly sets the clock-time when the server starts accepting connections. If a value is passed that evaluates to true, then the current clock time is marked as the starting time. In either case, the current value is returned. The clock-time is based on the internal time command of Perl, and thus is represented as an integer number of seconds since the system epoch. Generally, it is suitable for passing to either localtime or to the CWtime2iso8601 routine exported by the RPC::XML package.

timeout(INT)

You can call this method to set the timeout of new connections after they are received. This function returns the old timeout value. If you pass in no value then it will return the old value without modifying the current value. The default value is 10 seconds.

add_method(FILE | HASHREF | OBJECT)

add_proc(FILE | HASHREF | OBJECT)

This adds a new published method or procedure to the server object that invokes it. The new method may be specified in one of three ways: as a filename, a hash reference or an existing object (generally of either RPC::XML::Procedure or RPC::XML::Method classes). If passed as a hash reference, the following keys are expected:

name

The published (externally-visible) name for the method.

version

An optional version stamp. Not used internally, kept mainly for informative purposes.

hidden

If passed and evaluates to a CWtrue value, then the method should be hidden from any introspection API implementations. This parameter is optional, the default behavior being to make the method publically-visible.

code

A code reference to the actual Perl subroutine that handles this method. A symbolic reference is not accepted. The value can be passed either as a reference to an existing routine, or possibly as a closure. See How Methods are Called for the semantics the referenced subroutine must follow.

signature

A list reference of the signatures by which this routine may be invoked. Every method has at least one signature. Though less efficient for cases of exactly one signature, a list reference is always used for sake of consistency.

help

Optional documentation text for the method. This is the text that would be returned, for example, by a system.methodHelp call (providing the server has such an externally-visible method). If a file is passed, then it is expected to be in the XML-based format, described in the RPC::XML::Procedure manual (see RPC::XML::Procedure). If the name passed is not an absolute pathname, then the file will be searched for in any directories specified when the object was instantiated, then in the directory into which this module was installed, and finally in the current working directory. If the operation fails, the return value will be a non-reference, an error message. Otherwise, the return value is the object reference. The add_method and add_proc calls are essentialy identical unless called with hash references. Both files and objects contain the information that defines the type (method vs. procedure) of the funtionality to be added to the server. If add_method is called with a file that describes a procedure, the resulting addition to the server object will be a RPC::XML::Procedure object, not a method object. For more on the creation and manipulation of procedures and methods as objects, see RPC::XML::Procedure.

delete_method(N)

delete_proc(N)

Delete the named method or procedure from the calling object. Removes the entry from the internal table that the object maintains. If the method is shared across more than one server object (see share_methods), then the underlying object for it will only be destroyed when the last server object releases it. On error (such as no method by that name known), an error string is returned. The delete_proc call is identical, supplied for the sake of symmetry. Both calls return the matched object regardless of its underlying type.

list_methods

list_procs

This returns a list of the names of methods and procedures the server current has published. Note that the returned values are not the method objects, but rather the names by which they are externally known. The hidden status of a method is not consulted when this list is created; all methods and procedures known are listed. The list is not sorted in any specific order. The list_procs call is provided for symmetry. Both calls list all published routines on the calling server object, regardless of underlying type.

xpl_path([LISTREF])

Get and/or set the object-specific search path for CW*.xpl files (files that specify methods) that are specified in calls to add_method, above. If a list reference is passed, it is installed as the new path (each element of the list being one directory name to search). Regardless of argument, the current path is returned as a list reference. When a file is passed to add_method, the elements of this path are searched first, in order, before the installation directory or the current working directory are searched.

get_method(N)

get_proc(N)

Returns a reference to an object of the class RPC::XML::Method or RPC::XML::Procedure, which is the current binding for the published method NAME. If there is no such method known to the server, then CWundef is returned. The object is implemented as a hash, and has the same key and value pairs as for CWadd_method, above. Thus, the reference returned is suitable for passing back to CWadd_method. This facilitates temporary changes in what a published name maps to. Note that this is a referent to the object as stored on the server object itself, and thus changes to it could affect the behavior of the server. The get_proc interface is provided for symmetry.

server_loop(HASH)

Enters the connection-accept loop, which generally does not return. This is the CWaccept()-based loop of HTTP::Daemon if the object was created with an instance of that class as a part. Otherwise, this enters the run-loop of the Net::Server class. It listens for requests, and marshalls them out via the CWdispatch method described below. It answers HTTP-HEAD requests immediately (without counting them on the server statistics) and efficiently by using a cached HTTP::Response object. Because infinite loops requiring a CWHUP or CWKILL signal to terminate are generally in poor taste, the HTTP::Daemon side of this sets up a localized signal handler which causes an exit when triggered. By default, this is attached to the CWINT signal. If the Net::Server module is being used instead, it provides its own signal management. The arguments, if passed, are interpreted as a hash of key/value options (not a hash reference, please note). For HTTP::Daemon, only one is recognized:

signal

If passed, should be the traditional name for the signal that should be bound to the exit function. If desired, a reference to an array of signal names may be passed, in which case all signals will be given the same handler. The user is responsible for not passing the name of a non-existent signal, or one that cannot be caught. If the value of this argument is 0 (a CWfalse value) or the string CWCBNONECW, then the signal handler will not be installed, and the loop may only be broken out of by killing the running process (unless other arrangements are made within the application). The options that Net::Server responds to are detailed in the manual pages for that package. All options passed to CWserver_loop in this situation are passed unaltered to the CWrun() method in Net::Server.

dispatch(REQUEST)

This is the server method that actually manages the marshalling of an incoming request into an invocation of a Perl subroutine. The parameter passed in may be one of: a scalar containing the full XML text of the request, a scalar reference to such a string, or a pre-constructed RPC::XML::request object. Unless an object is passed, the text is parsed with any errors triggering an early exit. Once the object representation of the request is on hand, the parameter data is extracted, as is the method name itself. The call is sent along to the appropriate subroutine, and the results are collated into an object of the RPC::XML::response class, which is returned. Any non-reference return value should be presumed to be an error string. If the dispatched method encountered some sort of error, it will not be propagated upward here, but rather encoded as an object of the RPC::XML::fault class, and returned as the result of the dispatch. This distinguishes between server-centric errors, and general run-time errors.

add_default_methods([DETAILS])

This method adds all the default methods (those that are shipped with this extension) to the calling server object. The files are denoted by their CW*.xpl extension, and are installed into the same directory as this Server.pm file. The set of default methods are described below (see The Default Methods Provided). If any names are passed as a list of arguments to this call, then only those methods specified are actually loaded. If the CW*.xpl extension is absent on any of these names, then it is silently added for testing purposes. Note that the methods shipped with this package have file names without the leading CWstatus. part of the method name. If the very first element of the list of arguments is CWexcept (or CW-except), then the rest of the list is treated as a set of names to not load, while all others do get read. The Apache::RPC::Server module uses this to prevent the loading of the default CWsystem.status method while still loading all the rest of the defaults. (It then provides a more Apache-centric status method.) Note that there is no symmetric call in this case. The provided API is implemented as methods, and thus only this interface is provided.

add_methods_in_dir(DIR [, DETAILS])

add_procs_in_dir(DIR [, DETAILS])

This is exactly like add_default_methods above, save that the caller specifies which directory to scan for CW*.xpl files. In fact, the add_default_methods routine simply calls this routine with the installation directory as the first argument. The definition of the additional arguments is the same as above. add_procs_in_dir is provided for symmetry.

share_methods(SERVER, NAMES)

share_procs(SERVER, NAMES)

The calling server object shares the methods and/or procedures listed in NAMES with the source-server passed as the first object. The source must derive from this package in order for this operation to be permitted. At least one method must be specified, and all are specified by name (not by object refernce). Both objects will reference the same exact RPC::XML::Procedure (or Method, or derivative thereof) object in this case, meaning that call-statistics and the like will reflect the combined data. If one or more of the passed names are not present on the source server, an error message is returned and none are copied to the calling object. Alternately, one or more of the name parameters passed to this call may be regular-expression objects (the result of the qr operator). Any of these detected are applied against the list of all available methods known to the source server. All matching ones are inserted into the list (the list is pared for redundancies in any case). This allows for easier addition of whole classes such as those in the CWsystem.* name space (via CBqr/^system\./), for example. There is no substring matching provided. Names listed in the parameters to this routine must be either complete strings or regular expressions. The share_procs interface is provided for symmetry.

copy_methods(SERVER, NAMES)

copy_procs(SERVER, NAMES)

This behaves like the method share_methods above, with the exception that the calling object is given a clone of each method, rather than referencing the same exact method as the source server. The code reference part of the method is shared between the two, but all other data are copied (including a fresh copy of any list references used) into a completely new RPC::XML::Procedure (or derivative) object, using the CWclone() method from that class. Thus, while the calling object has the same methods available, and is re-using existing code in the Perl runtime, the method objects (and hence the statistics and such) are kept separate. As with the above, an error is flagged if one or more are not found. This routine also accepts regular-expression objects with the same behavior and limitations. Again, copy_procs is simply provided for symmetry.

Specifying the methods themselves can be a tricky undertaking. Some packages (in other languages) delegate a specific class to handling incoming requests. This works well, but it can lead to routines not intended for public availability to in fact be available. There are also issues around the access that the methods would then have to other resources within the same running system.

The approach taken by RPC::XML::Server (and the Apache::RPC::Server subclass of it) require that methods be explicitly published in one of the several ways provided. Methods may be added directly within code by using CWadd_method as described above, with full data provided for the code reference, signature list, etc. The CWadd_method technique can also be used with a file that conforms to a specific XML-based format (detailed in the manual page for the RPC::XML::Procedure class, see RPC::XML::Procedure). Entire directories of files may be added using CWadd_methods_in_dir, which merely reads the given directory for files that appear to be method definitions.

When a routine is called via the server dispatcher, it is called with the arguments that the client request passed. Depending on whether the routine is considered a procedure or a method, there may be an extra argument at the head of the list. The extra argument is present when the routine being dispatched is part of a RPC::XML::Method object. The extra argument is a reference to a RPC::XML::Server object (or a subclass thereof). This is derived from a hash reference, and will include two special keys:

method_name

This is the name by which the method was called in the client. Most of the time, this will probably be consistent for all calls to the server-side method. But it does not have to be, hence the passing of the value.

signature

This is the signature that was used, when dispatching. Perl has a liberal view of lists and scalars, so it is not always clear what arguments the client specifically has in mind when calling the method. The signature is an array reference containing one or more datatypes, each a simple string. The first of the datatypes specifies the expected return type. The remainder (if any) refer to the arguments themselves.

Note that by passing the server object reference first, method-classed routines are essentially expected to behave as actual methods of the server class, as opposed to ordinary functions. Of course, they can also discard the initial argument completely.

The routines should not make (excessive) use of global variables, for obvious reasons. When the routines are loaded from XPL files, the code is created as a closure that forces execution in the RPC::XML::Procedure package. If the code element of a procedure/method is passed in as a direct code reference by one of the other syntaxes allowed by the constructor, the package may well be different. Thus, routines should strive to be as localized as possible, independant of specific namespaces. If a group of routines are expected to work in close concert, each should explicitly set the namespace with a CWpackage declaration as the first statement within the routines themselves.

The following methods are provided with this package, and are the ones installed on newly-created server objects unless told not to. These are identified by their published names, as they are compiled internally as anonymous subroutines and thus cannot be called directly:

system.identity

Returns a string value identifying the server name, version, and possibly a capability level. Takes no arguments.

system.introspection

Returns a series of struct objects that give overview documentation of one or more of the published methods. It may be called with a string identifying a single routine, in which case the return value is a struct. It may be called with an array of string values, in which case an array of struct values, one per element in, is returned. Lastly, it may be called with no input parameters, in which case all published routines are documented. Note that routines may be configured to be hidden from such introspection queries.

system.listMethods

Returns a list of the published methods or a subset of them as an array of string values. If called with no parameters, returns all (non-hidden) method names. If called with a single string pattern, returns only those names that contain the string as a substring of their name (case-sensitive, and this is not a regular expression evaluation).

system.methodHelp

Takes either a single method name as a string, or a series of them as an array of string. The return value is the help text for the method, as either a string or array of string value. If the method(s) have no help text, the string will be null.

system.methodSignature

As above, but returns the signatures that the method accepts, as array of string representations. If only one method is requests via a string parameter, then the return value is the corresponding array. If the parameter in is an array, then the returned value will be an array of array of string.

system.multicall

This is a simple implementation of composite function calls in a single request. It takes an array of struct values. Each struct has at least a CWmethodName member, which provides the name of the method to call. If there is also a CWparams member, it refers to an array of the parameters that should be passed to the call.

system.status

Takes no arguments and returns a struct containing a number of system status values including (but not limited to) the current time on the server, the time the server was started (both of these are returned in both ISO 8601 and UNIX-style integer formats), number of requests dispatched, and some identifying information (hostname, port, etc.).

In addition, each of these has an accompanying help file in the CWmethods sub-directory of the distribution.

These methods are installed as CW*.xpl files, which are generated from files in the CWmethods directory of the distribution using the make_method tool (see make_method). The files there provide the Perl code that implements these, their help files and other information.

The RPC::XML::Server class now supports compressed messages, both incoming and outgoing. If a client indicates that it can understand compressed content, the server will use the Compress::Zlib (available from CPAN) module, if available, to compress any outgoing messages above a certain threshhold in size (the default threshhold is set to 4096 bytes). The following methods are all related to the compression support within the server class:

compress

Returns a false value if compression is not available to the server object. This is based on the availability of the Compress::Zlib module at start-up time, and cannot be changed.

compress_thresh([MIN_LIMIT])

Return or set the compression threshhold value. Messages smaller than this size in bytes will not be compressed, even when compression is available, to save on CPU resources. If a value is passed, it becomes the new limit and the old value is returned.

If the server anticipates handling large out-bound messages (for example, if the hosted code returns large Base64 values pre-encoded from file handles), the CWmessage_file_thresh and CWmessage_temp_dir settings may be used in a manner similar to RPC::XML::Client. Specifically, the threshhold is used to determine when a message should be spooled to a filehandle rather than made into an in-memory string (the RPC::XML::base64 type can use a filehandle, thus eliminating the need for the data to ever be completely in memory). An anonymous temporary file is used for these operations.

Note that the message size is checked before compression is applied, since the size of the compressed output cannot be known until the full message is examined. It is possible that a message will be spooled even if its compressed size is below the threshhold, if the uncompressed size exceeds the threshhold.

message_file_thresh

message_temp_dir

These methods may be used to retrieve or alter the values of the given keys as defined earlier for the CWnew method.

Unless explicitly stated otherwise, all methods return some type of reference on success, or an error string on failure. Non-reference return values should always be interpreted as errors unless otherwise noted.

This began as a reference implementation in which clarity of process and readability of the code took precedence over general efficiency. It is now being maintained as production code, but may still have parts that could be written more efficiently.

The XML-RPC standard is Copyright (c) 1998-2001, UserLand Software, Inc. See <http://www.xmlrpc.com> for more information about the XML-RPC specification. A helpful patch was sent in by Tino Wuensche to fix problems in the signal-setting and signal-catching code in server_loop().

This module is licensed under the terms of the Artistic License that covers Perl. See <http://www.opensource.org/licenses/artistic-license.php> for the license.

RPC::XML, RPC::XML::Client, RPC::XML::Parser

Randy J. Ray <rjray@blackperl.com>