mandelsoft / engine

Request engine

An Egine to process Flow Graphs described by Resource Manifests

This project provides a controller working on an object space, which processes a flow graph described by resource objects stored in the object space.

The graph is an instance of a graph metamodel, which is passed to the controller. The controller itself provides a metamodel used to describe graph metamodels.

A graph metamodel defines the node types (whose instances are stored in the object space) and the potential dependencies among instances of such types.

A concrete model with dedicated instances, their types and dependencies is then stored in the objectspace.

The processing is based on a model of shadow elements, which implement the concrete processing steps. They are wrapping persistent objects stored in the object space.

There are the user facing defining object types (external objects), which describe the desired state of the processing graph.

Every external object is typically shadowed by an internal (state) object, which keeps the processing state.

On top of those persistent objects, the programming model uses implementation objects, which must be provided by an implementation of the graph metamodel. Those objects are instantiated by the engine in memory and must implement the engine interface. They are used by the engine to provide standardized information, like the external desired state, or the dependency information. And they execute the payload related processing steps. The flow control is completely done by engine.

The processing of an internal object is divided into one or more processing phases. Every phase has the same interface towards the engine:

• provide current state
• accepting target state from a triggering external object
• providing dependency information according to current and target state
• providing a processing step (which might provide incremental progress)
• providing an output state after successful processing.

The engine then handles the dataflow among the processing graph defined by the current and target state of the phases and their dependencies. The output of input dependencies is provided to the processing, and if it is successful the output of the processing is again used as input for dependent phases.

Every external object has an assigned internal object (which is created by engine on demand), which handles the concrete processing. Therefore, changes on the external object trigger the processing of a phase of the internal object. Multiple phases might be provided by the internal object, which feature different dependencies.

Thereby, three kinds of relations between an external and an internal (state carrying) objects are supported:

1. Every external object is used to control the life-cycle of the internal one
2. An external object is controlled by a phase of an internal object.
3. A phase of an internal object may create other internal objects with dependencies of successor phases of the internal object.

Testing

The project includes a filesystem-based implementation of an objectspace. Additionally, a web server is included, which can be used to access and manipulate the objectspace.

It can be started with the engine command. By default, it is linked with an implemention of the sample metamodel provided by this project.

With the command ectl resource objects can be retrieved and applied.

Both commands can be installed with make install.

Example metamodel

The project come with an example metamodel describing a simple calculator.

It describes Value objects (case 1) usable as operands for Operator objects.

The operator manages an expression object responsible for the execution of the operation. This object is processed by a dedicated expression controller. The opertor just gather the operands, creates an expression object (case 3), waits for the result (case 2) and provides the output to described output value object (case 3).

The concrete implementation uses an expression controller, which now may create again sub processing graphs to calculate more complex expressions like A*B+C.

An example for a concrete model based on this metamodel is shown here:

Running the example

First, start the engine with

engine engine -d local/db -F ./ui -P /watch -c -D 1s

Adding the option -L debug will provide more logging output. With -D 1s it is possible to slow down the processing to observe the processing flow. With a web browser the URL http://localhost:8080/ui starts a simple visualization of the processing based on the phases can be seen.

The folder scenarios contains some example scenarios which can be deployed using

ectl apply -f *.yaml

With

ectl get -c 
ectl get Value

it is possible to display the involved objects and their state.