Add state support in fail scenario

Brief introduction

Example of scenario with a deployed instance with two elements:

user call undeploy command
first instance element success with UNDEPLOYED state
second instance element fail with DEPLOYED state
user call undeploy command again
first instance element does not know that the element is already UNDEPLOYED

Add support for the participant to understand what was the last state.

Allowed state from the participant perspective

Action	state	stChResult	Description
Prime	PRIMED	NO_ERROR	Prime is completed
Prime	COMMISSIONED	FAILED	Prime is failed
Deprime	COMMISSIONED	NO_ERROR	Deprime is completed
Deprime	DEPRIMED	FAILED	Deprime is failed

Action	deployState	lockState	stChResult	Description
Deploy	DEPLOYED		NO_ERROR	Deploy is completed
Deploy	UNDEPLOYED		FAILED	Deploy is failed
Undeploy	UNDEPLOYED		NO_ERROR	Undeploy is completed
Undeploy	DEPLOYED		FAILED	Undeploy is failed
Lock		LOCKED	NO_ERROR	Lock is completed
Lock		UNLOCKED	FAILED	Lock is failed
Unlock		UNLOCKED	NO_ERROR	Unlock is completed
Unlock		LOCKED	FAILED	Unlock is failed
Update	DEPLOYED		NO_ERROR	Update is completed
Update	DEPLOYED		FAILED	Update is failed
Migrate	DEPLOYED		NO_ERROR	Migrate is completed
Migrate	DEPLOYED		FAILED	Migrate is failed
Delete	DELETED		NO_ERROR	Delete is completed
Delete	UNDEPLOYED		FAILED	Delete is failed

Additional comments

It is important to make distinction between the state of the instance/element in the flow, and the state of the application involved. A deployed element means that a participant has completed a deploy action, and should not be confused with a deployed application.

There are scenarios where it make sense when a process is failed, all elements should repeat the action, but in other scenarios it could be different, it depends on the context. Anyway, a participant could implement idempotent operations.

The responsibility to know what it needs to be done is delegated to the participant.

When a participant cannot implement idempotent operations, in order to cover the scenario with failed process with more than one element involved, it has to know what was the previous state.

Solutions

A simple solution from participant side is that it can use "outProperties" to save information related to the deployment process.

"outProperties" is controlled by the participant side, and saved into ACM-r Db by status message (by sendAcElementInfo method). Each participant replica will be synchronized as well.

The examples below are suggestions.

Below a simple example:

    @Override
    public void deploy(CompositionElementDto compositionElement, InstanceElementDto instanceElement) throws PfModelException {

        if ("DEPLOYED".equals(instanceElement.outProperties().get("state"))) {
            // deploy process already done
            intermediaryApi.updateAutomationCompositionElementState(instanceElement.instanceId(), instanceElement.elementId(), DeployState.DEPLOYED, null, StateChangeResult.NO_ERROR, "Already Deployed");
            return;
        }
         
        // deployment process
        .......................................
        .......................................
        // end of the deployment process
        
        if (isDeploySuccess()) {
            instanceElement.outProperties().put("state", "DEPLOYED");
            intermediaryApi.sendAcElementInfo(instanceElement.instanceId(), instanceElement.elementId(), null, null, instanceElement.outProperties());
 
            intermediaryApi.updateAutomationCompositionElementState(instanceElement.instanceId(), instanceElement.elementId(), DeployState.DEPLOYED, null, StateChangeResult.NO_ERROR, "Deployed");
        } else {
            instanceElement.outProperties().put("state", "UNDEPLOYED");
            intermediaryApi.sendAcElementInfo(instanceElement.instanceId(), instanceElement.elementId(), null, null, instanceElement.outProperties());
 
            intermediaryApi.updateAutomationCompositionElementState(instanceElement.instanceId(), instanceElement.elementId(), DeployState.UNDEPLOYED, null, StateChangeResult.FAILED, "Deploy failed!");
        } 
    }


    @Override
    public void undeploy(CompositionElementDto compositionElement, InstanceElementDto instanceElement) throws PfModelException {

        if ("DEPLOYED".equals(instanceElement.outProperties().get("state"))) {
            // undeploy process already done

            intermediaryApi.updateAutomationCompositionElementState(instanceElement.instanceId(), instanceElement.elementId(), DeployState.UNDEPLOYED, null, StateChangeResult.NO_ERROR, "Already Undeployed");
            return;
        }

        // undeployment process
        .......................................
        .......................................
        // end of the undeployment process        
        
        if (isUndeploySuccess()) {
             instanceElement.outProperties().put("state", "UNDEPLOYED");
            intermediaryApi.sendAcElementInfo(instanceElement.instanceId(), instanceElement.elementId(), null, null, instanceElement.outProperties());

            intermediaryApi.updateAutomationCompositionElementState(instanceElement.instanceId(), instanceElement.elementId(), DeployState.UNDEPLOYED, null, StateChangeResult.NO_ERROR, "Undeployed");
        } else {
            instanceElement.outProperties().put("state", "DEPLOYED");
            intermediaryApi.sendAcElementInfo(instanceElement.instanceId(), instanceElement.elementId(), null, null, instanceElement.outProperties());

            intermediaryApi.updateAutomationCompositionElementState(instanceElement.instanceId(), instanceElement.elementId(), DeployState.DEPLOYED, null, StateChangeResult.FAILED, "Undeploy failed!");
        }
    }

Below other suggestion when a process is failed, all elements can repeat the action and do some clean up

    @Override
    public void deploy(CompositionElementDto compositionElement, InstanceElementDto instanceElement) throws PfModelException {

        var state = instanceElement.outProperties().get("state");
        if ("DEPLOYED".equals(state)) {
            // clean up deployment
        
        } else if ("DEPLOYING".equals(state) || "UNDEPLOYING".equals(state)) {
            // check and clean up

        }

        // deployment process
        instanceElement.outProperties().put("state", "DEPLOYING");
        intermediaryApi.sendAcElementInfo(instanceElement.instanceId(), instanceElement.elementId(), null, null, instanceElement.outProperties()); 
        
        .......................................
        .......................................
        .......................................
        // end of the deployment process        
        
         if (isDeploySuccess()) {
            instanceElement.outProperties().put("state", "DEPLOYED");
            intermediaryApi.sendAcElementInfo(instanceElement.instanceId(), instanceElement.elementId(), null, null, instanceElement.outProperties());
 
            intermediaryApi.updateAutomationCompositionElementState(instanceElement.instanceId(), instanceElement.elementId(), DeployState.DEPLOYED, null, StateChangeResult.NO_ERROR, "Deployed");
        } else {
            instanceElement.outProperties().put("state", "UNDEPLOYED");
            intermediaryApi.sendAcElementInfo(instanceElement.instanceId(), instanceElement.elementId(), null, null, instanceElement.outProperties());
 
            intermediaryApi.updateAutomationCompositionElementState(instanceElement.instanceId(), instanceElement.elementId(), DeployState.UNDEPLOYED, null, StateChangeResult.FAILED, "Deploy failed!");
        } 
     }


    @Override
    public void undeploy(CompositionElementDto compositionElement, InstanceElementDto instanceElement) throws PfModelException {

         var state = instanceElement.outProperties().get("state");
        if ("UNDEPLOYED".equals(state)) {
            // clean up undeployment
        
        } else if ("DEPLOYING".equals(state) || "UNDEPLOYING".equals(state)) {
            // check and clean up

        }
 
        // undeployment process
 
        instanceElement.outProperties().put("state", "UNDEPLOYING");
        intermediaryApi.sendAcElementInfo(instanceElement.instanceId(), instanceElement.elementId(), null, null, instanceElement.outProperties()); 

        .......................................
        .......................................
        .......................................
        .......................................
        // end of the undeployment process        
        
         if (isUndeploySuccess()) {
            instanceElement.outProperties().put("state", "UNDEPLOYED");
            intermediaryApi.sendAcElementInfo(instanceElement.instanceId(), instanceElement.elementId(), null, null, instanceElement.outProperties());

            intermediaryApi.updateAutomationCompositionElementState(instanceElement.instanceId(), instanceElement.elementId(), DeployState.UNDEPLOYED, null, StateChangeResult.NO_ERROR, "Undeployed");
        } else {
            instanceElement.outProperties().put("state", "DEPLOYED");
            intermediaryApi.sendAcElementInfo(instanceElement.instanceId(), instanceElement.elementId(), null, null, instanceElement.outProperties());

            intermediaryApi.updateAutomationCompositionElementState(instanceElement.instanceId(), instanceElement.elementId(), DeployState.DEPLOYED, null, StateChangeResult.FAILED, "Undeploy failed!");
        }
    }

Final consideration: In all suggestions before we have used labels as "DEPLOY", "UNDEPLOY", "DEPLOYING", "UNDEPLOYING" but the developer can change them as better suit with the context of the participant.