Data Definition Language
As with other remote procedure invocation systems MCollective has a DDL that defines what remote methods are available, what inputs they take and what outputs they generate.
In addition to the usual procedure definitions we also keep meta data about author, versions, license and other key data points.
The DDL is used in various scenarios:
- The user can access it in the form of a human readable help page
- User interfaces can access it in a way that facilitate auto generation of user interfaces
- The RPC client auto configures and use appropriate timeouts in waiting for responses
- Before sending a call over the network inputs get validated so we do not send unexpected data to remote nodes.
- Module repositories can use the meta data to display a standard view of available modules to assist a user in picking the right ones.
- The server will validate incoming requests prior to sending it to agents
We’ve created a screencast showing the capabilities of the DDL that might help give you a better overview.
NOTE: As of version 2.1.1 the DDL is required on all servers before an agent will be activated
We’ll start with a few examples as I think it’s pretty simple what they do, and later on show what other permutations are allowed for defining inputs and outputs.
A helper agent called rpcutil is included that helps you gather stats, inventory etc about the running daemon. This helper has a full DDL included, see the plugins dir for this agent.
The typical service agent is a good example, it has various actions that all more or less take the same input. All but status would have almost identical language.
First we need to define the meta data for the agent itself:
It’s fairly obvious what these all do, :timeout is how long the MCollective daemon will let the threads run.
As of MCollective 2.1.2 you can indicate which is the lowest version of MCollective needed to use a plugin. Plugins that do not meet the requirement can not be used.
You should add this right after the metadata section in the DDL
Actions, Input and Output
Defining inputs and outputs is the hardest part, below first the status action:
As you see we can define all the major components of input and output parameters. :type can be one of various values and each will have different parameters, more on that later.
As of version 2.1.1 the outputs can define a default value. For agents the reply structures are pre-populated with all the defined outputs, if no default is supplied a default of nil will be set.
As of version 2.3.1 the inputs can also define default values, this is only processed and applied for non optional inputs.
By default mcollective only show data from actions that failed, the display line above tells it to always show the results. Possible values are :ok, :failed (the default behavior) and :always.
Finally the service agent has 3 almost identical actions - start, stop and restart - below we use a simple loop to define them all in one go.
All of this code just goes into a file, no special class or module bits needed, just save it as service.ddl in the same location as the service.rb.
Importantly you do not need to have the service.rb on a machine to use the DDL, this means on machines that are just used for running client programs you can just drop the .ddl files into the agents directory.
You can view a human readable version of this using mco plugin doc <agent> command:
The input block has a mandatory :optional field, when true it would be ok if a client attempts to call the agent without this input supplied. If it is supplied though it will be validated.
Types of Input
As you see above the input block has :type option, types can be :string, :list, :boolean, :integer, :float or :number
The string type validates initially that the input is infact a String, then it validates the length of the input and finally matches the supplied Regular Expression.
Both :validation and :maxlength are required arguments for the string type of input.
If you want to allow unlimited length text you can make :maxlength => 0 but use this with care.
As of version 2.2.0 a new plugin type called Validator Plugins exist that allow you to supply your own validations for :string types.
List types provide a list of valid options and only those will be allowed, see an example below:
In user interfaces this might be displayed as a drop down list selector or another kind of menu.
The value input should be either true or false actual boolean values. This feature was introduced in version 0.4.9.
The value input should be an integer number like 1 or 100 but not 1.1. This feature was introduced in version 1.3.2
The value input should be a floating point number like 1.0 but not 1. This feature was introduced in version 1.3.2
The value input should be an integer or a floating point number. This feature was introduced in version 1.3.2
The value input can be any type, this allows you to send rich objects like arrays of hashes around, it effectively disables validation of the type of input.
The :any type is deprecated and will be removed after version 2.2.x.
Accessing the DDL
While programming client applications or web apps you can gain access to the DDL for any agent in several ways:
This will produce the text help output from the above example, you can supply any ERB template to format the output however you want.
You can also access the data structures directly:
The ddl object is also available on any rpcclient:
In the case of accessing it through the service as in this example, if there was no DDL file on the machine for the service agent you’d get a nil back from the ddl accessor.
As mentioned earlier the client does automatic input validation using the DDL, if validation fails you will get an MCollective::DDLValidationError exception thrown with an appropriate message.