Project to rewrite parts of Horde_Rpc and the Horde Core Registry inter-app API for a new major release
List any tickets on http://bugs.horde.org/ that cover this issue or are relevant to it.
Main Design Goals:
- Facilitate mapping of free form URIs to API calls (for REST)
- Facilitate customizing/amending APIs by administrator/downstream
- Allow passing PHP Objects when using the API internally -> Need a defined way to fold them to array/scalar representations on RPC
- Allow separate authentication mechanisms for external calls (REST, RPC) like using API keys, even limiting them.
- Allow unauthenticated usage of select RPC API calls.
- Ability for administrators to turn off remote access to some APIs
- Avoid boilerplate inside Api method implementation
- Little or no setup for default case
- Allow but don't require versioning
- keep code upgrade costs from H5 to H6 reasonable
- Allow routing API calls to external horde instances
An API is a named collection of methods
H5 registry allows to route an API to an app and to override this decision for certain methods
For example, ingo is the default target for "filter" api and the defined target for certain methods of the "mail" api, which defaults to imp
'provides' => array(
However, currently an App cannot provide separate implementations for the same method name in different APIs. Calling filter/newEmailFilter would yield the mail/newEmailFilter method due to all methods being part of one collective Api.php
The current API also does not allow named parameters, even if some RPC mechanisms would deal with it
If an API provides an evolving protocol, discerning versions and supporting older input and output formats would be the method implementation's task
The registry allows limited runtime introspection (hasMethod, listMethods, listApis) but no hints on input/output formats (parameter list and type)
A simple boilerplate router should be part of Horde_Rpc to facilitate unit testing
A Horde-specific router Horde_Core_Rpc_Router should be part of Horde_Core and interact with registry:
Check if api exists (for RPC access / individual RPC type)
Check if method exists (for RPC access / individual RPC type)
List available apis (for RPC access / individual RPC type) - should be cached
List available methods (for RPC access / individual RPC type) - should be cached
Find right app to call a specific method
Find the right runner and configuration for a set of rpc type, api, method
Most RPC types have a fixed request pattern allowing to easily find out the API and method to look for - we can easily delegate this to the registry/internal API and then check if the found result is exposed by the app
Check for class $App_Api_$Api_$Type if it's explicitly allowed for this RPC type
Check for class $App_Api_$Api_Rpc (inherits from Horde_Core_Api_Rpc) if it's generally allowed for most RPC types
First check for custom overrides $App_Api_$Api_$TypeLocal and $App_Api_$Api_RpcLocal ? Does this hit performance badly?
Don't expose this internal API via RPC at all if no such class exists
REST does not have a fixed request pattern. The Router needs to know all REST patterns of all APPs before it can even guess if some requested api method exists anywhere. Therefore, REST needs an own Horde_Core_Rpc_Router_Rest and the RPC Endpoint rpc.php needs to be aware of that fact
For each Api/Provider combination in registry
Check for class $App_Api_$Api_Rest
Don't expose this internal API via Rest at all if no such class exists
Each identified Rest request must be checked against this list of known REST patterns by Horde_Routes_Mapper.
Each identified Rest request must have a defined answer format to apply to the internal api result (let's check if there is a reasonable per-verb default for most cases when implementing)
Horde_Core_Api_Rpc (and descendants)
Call internal api $App_Api_$Api to actually run a command
should check what returns for PHP objects and reduce them to arrays (__sleep or Serializable)
allow plugging separate authentication, accounting ...
may have to be amended by specific classes or config to support some RPC types (REST)
In Question: Does it make sense to also wrap internal calls into a Runner_Internal ?
The Internal runner simply moves setting up the execution environment and calling the actual class and method from the registry class.
As of now, no additional functionality is considered but it may be handy for debugging.
Accounting is optional and will probably only be implemented if somebody is interested. Accounting defines a set of limits (value per timespan) and stores usage timestamps in a backend (user, timestamp, action). If the number of usages inside a sliding window hits the limit, the request is denied.
Asking for a past window (like first of month, first of next month) allows for billing/reporting, if this is of any interest.
Implementation could use horde_histories, horde_locks (no billing due to cleanup) or a text file backend (probably inefficient).
The accounting lib can probably be reused for other things.
For internal app calls, git mv $app/lib/Api.php to $app/lib/Api/$Api.php and change class name. Move overrides into a separate class. Done.
For most external calls, additionally extend to-be-written Horde_Core_Api_Rpc into $App_Api_$Api_Rpc - probably the defaults will be just fine for upgrade.
Protection modules may be combined, probably at first we only want Horde Auth (like Login, for ActiveSync, DAV) and Extra Auth (API keys or SSL)
Horde Default Auth
Need Permission (implies any auth)
Extra Auth (how to map to Horde Users for permissions?)
Per-User global accounting
Per-User Per-Api accounting
Per-User Per-Method accounting
Decide which RPC Types are generally enabled. Default?
Decide which RPC Types use Horde Auth, Separate Auth or both
Set up accounting system at all (yes/no)
Set up global per-user accounting (yes/no)
Set up anonymous accounting (yes/no)
Upstream provides defaults on which RPC methods are available and how they are protected (multiple available).
Admin / Config may override, empty config uses defaults
Back to the Project List