Entities

1. Assets

Assets are containers for metadata, they do not contain the actual bits and bytes. Say you want to offer a file to the dataspace, that is physically located in an S3 bucket, then the corresponding Asset would contain metadata about it, such as the content type, file size, etc. In addition, it could contain private properties, for when you want to store properties on the asset, which you do not want to expose to the dataspace. Private properties will get ignored when serializing assets out over DSP.

A very simplistic Asset could look like this:

{
  "@context": {
    "edc": "https://w3id.org/edc/v0.0.1/ns/"
  },
  "@id": "79d9c360-476b-47e8-8925-0ffbeba5aec2",
  "properties": {
    "somePublicProp": "a very interesting value"
  },
  "privateProperties": {
    "secretKey": "this is secret information, never tell it to the dataspace!"
  },
  "dataAddress": {
    "type": "HttpData",
    "baseUrl": "http://localhost:8080/test"
  }
}

The Asset also contains a DataAddress object, which can be understood as a “pointer into the physical world”. It contains information about where the asset is physically located. This could be a HTTP URL, or a complex object. In the S3 example, that DataAddress might contain the bucket name, region and potentially other information. Notice that the schema of the DataAddress will depend on where the data is physically located, for instance a HttpDataAddress has different properties from an S3 DataAddress. More precisely, Assets and DataAddresses are schemaless, so there is no schema enforcement beyond a very basic validation. Read this document to learn about plugging in custom validators.

A few things must be noted. First, while there isn’t a strict requirement for the @id to be a UUID, we highly recommend using the JDK UUID implementation.

Second, never store access credentials such as passwords, tokens, keys etc. in the dataAddress or even the privateProperties object. While the latter does not get serialized over DSP, both properties are persisted in the database. Always use a HSM to store the credential, and hold a reference to the secret in the DataAddress. Checkout the best practices for details.

By design, Assets are extensible, so users can store any metadata they want in it. For example, the properties object could contain a simple string value, or it could be a complex object, following some custom schema. Be aware, that unless specified otherwise, all properties are put under the edc namespace by default. There are some “well-known” properties in the edc namespace: id, description, version, name, contenttype.

Here is an example of how an Asset with a custom property following a custom namespace would look like:

{
  "@context": {
    "edc": "https://w3id.org/edc/v0.0.1/ns/",
    "sw": "http://w3id.org/starwars/v0.0.1/ns/"
  },
  "@id": "79d9c360-476b-47e8-8925-0ffbeba5aec2",
  "properties": {
    "faction": "Galactic Imperium",
    "person": {
      "name": "Darth Vader",
      "webpage": "https://death.star"
    }
  }
}

(assuming the sw context contains appropriate definitions for faction and person).

Remember that upon ingress through the Management API, all JSON-LD objects get expanded, and the control plane only operates on expanded JSON-LD objects. The Asset above would look like this:

[
  {
    "@id": "79d9c360-476b-47e8-8925-0ffbeba5aec2",
    "https://w3id.org/edc/v0.0.1/ns/properties": [
      {
        "https://w3id.org/starwars/v0.0.1/ns/faction": [
          {
            "@value": "Galactic Imperium"
          }
        ],
        "http://w3id.org/starwars/v0.0.1/ns/person": [
          {
            "http://w3id.org/starwars/v0.0.1/ns/name": [
              {
                "@value": "Darth Vader"
              }
            ],
            "http://w3id.org/starwars/v0.0.1/ns/webpage": [
              {
                "@value": "https://death.star"
              }
            ]
          }
        ]
      }
    ]
  }
]

This is important to keep in mind, because it means that Assets get persisted in their expanded form, and operations performed on them (e.g. querying) in the control plane must also be done on the expanded form. For example, a query targeting the sw:faction field from the example above would look like this:

{
  "https://w3id.org/edc/v0.0.1/ns/filterExpression": [
    {
      "https://w3id.org/edc/v0.0.1/ns/operandLeft": [
        {
          "@value": "https://w3id.org/starwars/v0.0.1/ns/faction"
        }
      ],
      "https://w3id.org/edc/v0.0.1/ns/operator": [
        {
          "@value": "="
        }
      ],
      "https://w3id.org/edc/v0.0.1/ns/operandRight": [
        {
          "@value": "Galactic Imperium"
        }
      ]
    }
  ]
}

2. Policies

Policies are the EDC way of expressing that certain conditions may, must or must not be satisfied in certain situations. Policies are used to express what requirements a subject (e.g. a communication partner) must fulfill in order to be able to perform an action. For example, that the communication partner must be headquartered in the European Union.

Policies are ODRL serialized as JSON-LD. Thus, our previous example would look like this:

{
  "@context": {
    "edc": "https://w3id.org/edc/v0.0.1/ns/"
  },
  "@type": "PolicyDefinition",
  "policy": {
    "@context": "http://www.w3.org/ns/odrl.jsonld",
    "@type": "Set",
    "duty": [
      {
        "target": "http://example.com/asset:12345",
        "action": "use",
        "constraint": {
          "leftOperand": "headquarter_location",
          "operator": "eq",
          "rightOperand": "EU"
        }
      }
    ]
  }
}

The duty object expresses the semantics of the constraint. It is a specialization of rule, which expresses either a MUST (duty), MAY (permission) or MUST NOT (prohibition) relation. The action expresses the type of action for which the rule is intended. Acceptable values for action are defined here, but in EDC you’ll exclusively encounter "use".

The constraint object expresses logical relationship of a key (leftOperand), the value (righOperand) and the operator. Multiple constraints can be linked with logical operators, see advanced policy concepts. The leftOperand and rightOperand are completely arbitrary, only the operator is limited to the following possible values: eq, neq, gt, geq, lt, leq, in, hasPart, isA, isAllOf, isAnyOf, isNoneOf.

Please note that not all operators are always allowed, for example headquarter_location lt EU is nonsensical and should result in an evaluation error, whereas headquarter_location isAnOf [EU, US] would be valid. Whether an operator is valid is solely defined by the policy evaluation function, supplying an invalid operator should raise an exception.

2.1 Policy vs PolicyDefinition

In EDC we have two general use cases under which we handle and persist policies:

  1. for use in contract definitions
  2. during contract negotiations

In the first case policies are ODRL objects and thus must have a uid property. They are typically used in contract definitions.

Side note: the ODRL context available at http://www.w3.org/ns/odrl.jsonld simply defines uid as an alias to the @id property. This means, whether we use uid or @id doesn’t matter, both expand to the same property @id.

However in the second case we are dealing with DCAT objects, that have no concept of Offers, Policies or Assets. Rather, their vocabulary includes Datasets, Dataservices etc. So when deserializing those DCAT objects there is no way to reconstruct Policy#uid, because the JSON-LD structure does not contain it.

To account for this, we defined the Policy class as value object that contains rules and other properties. In addition, we have a PolicyDefinition class, which contains a Policy and an id property, which makes it an entity.

2.2 Policy scopes and bindings

A policy scope is the “situation”, in which a policy is evaluated. For example, a policy may need to be evaluated when a contract negotiation is attempted. To do that, EDC defines certain points in the code called “scopes” to which policies are bound. These policy scopes (sometimes called policy evaluation points) are static, injecting/adding additional scopes is not possible. Currently, the following scopes are defined:

  • contract.negotiation: evaluated upon initial contract offer. Ensures that the consumer fulfills the contract policy.
  • transfer.process: evaluated before starting a transfer process to ensure that the policy of the contract agreement is fulfilled. One example would be contract expiry.
  • catalog: evaluated when the catalog for a particular participant agent is generated. Decides whether the participant has the asset in their catalog.
  • request.contract.negotiation: evaluated on every request during contract negotiation between two control plane runtimes. Not relevant for end users.
  • request.transfer.process: evaluated on every request during transfer establishment between two control plane runtimes. Not relevant for end users.
  • request.catalog: evaluated upon an incoming catalog request. Not relevant for end users.
  • provision.manifest.verify: evaluated during the precondition check for resource provisioning. Only relevant in advanced use cases.

A policy scope is a string that is used for two purposes:

  1. binding a scope to a rule type: implement filtering based on the action or the leftOperand of a policy. This determines for every rule inside a policy whether it should be evaluated in the given scope. In other words, it determines if a rule should be evaluated.
  2. binding a policy evaluation function to a scope: if a policy is determined to be “in scope” by the previous step, the policy engine invokes the evaluation function that was bound to the scope to evaluate if the policy is fulfilled. In other words, it determines (implements) how a rule should be evaluated.

2.3 Policy evaluation functions

If policies are a formalized declaration of requirements, policy evaluation functions are the means to evaluate those requirements. They are pieces of Java code executed at runtime. A policy on its own only expresses the requirement, but in order to enforce it, we need to run policy evaluation functions.

Upon evaluation, they receive the operator, the rightOperand (or rightValue), the rule, and the PolicyContext. A simple evaluation function that asserts the headquarters policy mentioned in the example above could look similar to this:

import org.eclipse.edc.policy.engine.spi.AtomicConstraintFunction;

public class HeadquarterFunction implements AtomicConstraintFunction<Duty> {
    public boolean evaluate(Operator operator, Object rightValue, Permission rule, PolicyContext context) {
        if (!(rightValue instanceof String)) {
            context.reportProblem("Right-value expected to be String but was " + rightValue.getClass());
            return false;
        }
        if (operator != Operator.EQ) {
            context.reportProblem("Invalid operator, only EQ is allowed!");
            return false;
        }

        var participant = context.getContextData(ParticipantAgent.class);
        var participantLocation = extractLocationClaim(participant); // EU, US, etc.
        return participantLocation != null && rightValue.equalsIgnoreCase(participantLocation);
    }
}

This particular evaluation function only accepts eq as operator, and only accepts scalars as rightValue, no list types.

The ParticipantAgent is a representation of the communication counterparty that contains a set of verified claims. In the example, extractLocationClaim() would look for a claim that contains the location of the agent and return it as string. This can get quite complex, for example, the claim could contain geo-coordinates, and the evaluation function would have to perform inverse address geocoding.

Other policies may require other context data than the participant’s location, for example an exact timestamp, or may even need a lookup in some third party system such as a customer database.

The same policy can be evaluated by different evaluation functions, if they are meaningful in different contexts (scopes).

NB: to write evaluation code for policies, implement the org.eclipse.edc.policy.engine.spi.AtomicConstraintFunction interface. There is a second interface with the same name, but that is only used for internal use in the PolicyEvaluationEngine.

2.4 Example: binding an evaluation function

As we’ve learned, for a policy to be evaluated at certain points, we need to create a policy (duh!), bind the policy to a scope, create a policy evaluation function, and we need to bind the function to the same scope. The standard way of registering and binding policies is done in an extension. For example, here we configure our HeadquarterFunction so that it evaluates our headquarter_location function whenever someone tries to negotiate a contract:

public class HeadquarterPolicyExtension implements ServiceExtension {

    @Inject
    private RuleBindingRegistry ruleBindingRegistry;

    @Inject
    private PolicyEngine policyEngine;

    private static final String HEADQUARTER_POLICY_KEY = "headquarter_location";

    @Override
    public void initialize() {
        // bind the policy to the scope
        ruleBindingRegistry.bind(HEADQUARTER_POLICY_KEY, NEGOTIATION_SCOPE);
        // create the function object
        var function = new HeadquarterFunction();
        // bind the function to the scope
        policyEngine.registerFunction(NEGOTIATION_SCOPE, Duty.class, HEADQUARTER_POLICY_KEY, function);
    }
}

The code does two things: it binds the function key (= the leftOperand) to the negotiation scope, which means that the policy is “relevant” in that scope. Further, it binds the evaluation function to the same scope, which means the policy engine “finds” the function and executes it in the negotiation scope.

This example assumes, a policy object exists in the system, that has a leftOperand = headquarter_location. For details on how to create policies, please check out the OpenAPI documentation.

2.5 Advanced policy concepts

2.5.1 Pre- and Post-Evaluators

Pre- and post-validators are functions that are executed before and after the actual policy evaluation, respectively. They can be used to perform preliminary evaluation of a policy or to enrich the PolicyContext. For example, EDC uses pre-validators to inject DCP scope strings using dedicated ScopeExtractor objects.

2.5.2 Dynamic functions

These are very similar to AtomicConstraintFunctions, with one significant difference: they also receive the left-operand as function parameter. This is useful when the function cannot be bound to a left-operand of a policy, because the left-operand is not known in advance.

Let’s revisit our headquarter policy from earlier and change it a little:

{
  "@context": {
    "edc": "https://w3id.org/edc/v0.0.1/ns/"
  },
  "@type": "PolicyDefinition",
  "policy": {
    "@context": "http://www.w3.org/ns/odrl.jsonld",
    "@type": "Set",
    "duty": [
      {
        "target": "http://example.com/asset:12345",
        "action": "use",
        "constraint": {
          "or": [
            {
              "leftOperand": "headquarter.location",
              "operator": "eq",
              "rightOperand": "EU"
            },
            {
              "leftOperand": "headerquarter.numEmployees",
              "operator": "gt",
              "rightOperand": 5000
            }
          ]
        }
      }
    ]
  }
}

This means two things. One, our policy has changed its semantics: now we require the headquarter to be in the EU, or to have more than 5000 employees.

2.6 Bundled policy functions

2.6.1 Contract expiration function

3. Contract definitions

Contract definitions are how assets and policies are linked together. It is EDC’s way of expressing which policies are in effect for an asset. So when an asset (or several assets) are offered in the dataspace, a contract definition is used to express under what conditions they are offered. Those conditions are comprised of a contract policy and an access policy. The access policy determines, whether a participant will even get the offer, and the contract policy determines whether they can negotiate a contract for it. Those policies are referenced by ID, but foreign-key constrainta are not enforced. This means that contract definitions can be created ahead of time.

It is important to note that contract definitions are implementation details (i.e. internal objects), which means they never leave the realm of the provider, and they are never sent to the consumer via DSP.

  • access policy: determines whether a particular consumer is offered an asset when making a catalog request. For example, we may want to restrict certain assets such that only consumers within a particular geography can see them. Consumers outside that geography wouldn’t even have them in their catalog.
  • contract policy: determines the conditions for initiating a contract negotiation for a particular asset. Note that this only guarantees the successful initiation of a contract negotiation, it does not automatically guarantee the successful conclusion of it!

Contract definitions also contain an assetsSelector. THat is a query expression that defines all the assets that are included in the definition, like an SQL SELECT statement. With that it is possible to configure the same set of conditions (= access policy and contract policy) for a multitude of assets.

Please note that creating an assetSelector may require knowledge about the shape of an Asset and can get complex fairly quickly, so be sure to read the chapter about querying.

Here is an example of a contract definition, that defines an access policy and a contract policy for assets id1, id2 and id3 that must contain the "foo" : "bar" property.

{
  "@context": {
    "edc": "https://w3id.org/edc/v0.0.1/ns/"
  },
  "@type": "https://w3id.org/edc/v0.0.1/ns/ContractDefinition",
  "@id": "test-id",
  "edc:accessPolicyId": "access-policy-1234",
  "edc:contractPolicyId": "contract-policy-5678",
  "edc:assetsSelector": [
    {
      "@type": "https://w3id.org/edc/v0.0.1/ns/Criterion",
      "edc:operandLeft": "id",
      "edc:operator": "in",
      "edc:operandRight": [
        "id1",
        "id2",
        "id3"
      ]
    },
    {
      "@type": "https://w3id.org/edc/v0.0.1/ns/Criterion",
      "edc:operandLeft": "foo",
      "edc:operator": "=",
      "edc:operandRight": "bar"
    }
  ]
}

The sample expresses that a set of assets identified by their ID be made available under the access policy access-policy-1234 and contract policy contract-policy-5678, if they contain a property "foo" : "bar".

Note that asset selector expressions are always logically conjoined using an “AND” operation.

4. Contract negotiations

If a connector fulfills the contract policy, it may initiate the negotiation of a contract for a particular asset. During that negotiation, both parties can send offers and counter-offers that can contain altered terms (= policy) as any human would in a negotiation, and the counter-party may accept or reject them.

Contract negotiations have a few key aspects:

  • they target one asset
  • they take place between a provider and a consumer connector
  • they cannot be changed by the user directly
  • users can only decline, terminate or cancel them

As a side note it is also important to note that contract offers are ephemeral objects as they are generated on-the-fly for a particular participant, and they are never persisted in a database and thus cannot be queried through any API.

Contract negotiations are asynchronous in nature. That means after initiating them, they become (potentially long-running) stateful processes that are advanced by an internal state machine. The current state of the negotiation can be queried and altered through the management API.

Here’s a diagram of the state machine applied to contract negotiations:

Contract Negotiation State Machine

A contract negotiation can be initiated from the consumer side by sending a ContractRequest to the connector management API.

{
  "@context": {
    "@vocab": "https://w3id.org/edc/v0.0.1/ns/"
  },
  "@type": "ContractRequest",
  "counterPartyAddress": "http://provider-address",
  "protocol": "dataspace-protocol-http",
  "policy": {
    "@context": "http://www.w3.org/ns/odrl.jsonld",
    "@type": "odrl:Offer",
    "@id": "offer-id",
    "assigner": "providerId",
    "permission": [],
    "prohibition": [],
    "obligation": [],
    "target": "assetId"
  },
  "callbackAddresses": [
    {
      "transactional": false,
      "uri": "http://callback/url",
      "events": [
        "contract.negotiation"
      ],
      "authKey": "auth-key",
      "authCodeId": "auth-code-id"
    }
  ]
}

The counterPartyAddress is the address where to send the ContractRequestMessage via the specified protocol ( currently dataspace-protocol-http)

The policy should hold the same policy associated to the data offering chosen from the catalog, plus two additional properties:

  • assigner the providers participantId
  • target the asset (dataset) ID

In addition, the (optional) callbackAddresses array can be used to get notified about state changes of the negotiation. Read more on callbacks in the section about events and callbacks.

Note: if the policy sent by the consumer differs from the one expressed by the provider, the contract negotiation will fail and transition to a TERMINATED state.

5. Contract agreements

Once a contract negotiation is successfully concluded (i.e. it reaches the FINALIZED state), it “turns into” a contract agreement. It is always the provider connector that gives the final approval. Contract agreements are immutable objects that contain the final, agreed-on policy, the ID of the asset that the contract was negotiated for, the IDs of the negotiation parties and the exact signing date.

Note that in future iterations contracts will be cryptographically signed to further support the need for immutability and non-repudiation.

Like contract definitions, contract agreements are entities that only exist within the bounds of a connector.

About terminating contracts: once a contract negotiation has reached a terminal state TERMINATED or FINALIZED, it becomes immutable. This could be compared to not being able to scratch a signature off a physical paper contract. Cancelling or terminating a contract is therefor handled through other channels like eventing systems. The semantics of cancelling a contract are highly individual to each dataspace and may even bring legal side effects, so EDC cannot make an assumption here.

6. Catalog

The catalog contains the “data offerings” of a connector and one or multiple service endpoints to initiate a negotiation for those offerings.

Every data offering is represented by a Dataset object which contains a policy and one or multiple Distribution objects. A Distribution should be understood as a variant or representation of the Dataset. For instance, if a file is accessible via multiple transmission channels from a provider (HTTP and FTP), then each of those channels would be represented as a Distribution. Another example would be image assets that are available in different file formats (PNG, TIFF, JPEG).

A DataService object specifies the endpoint where contract negotiations and transfers are accepted by the provider. In practice, this will be the DSP endpoint of the connector.

The following example shows an HTTP response to a catalog request, that contains one offer that is available via two channels HttpData-PUSH and HttpData-PULL.

catalog example
{
  "@id": "567bf428-81d0-442b-bdc8-437ed46592c9",
  "@type": "dcat:Catalog",
  "dcat:dataset": [
    {
      "@id": "asset-2",
      "@type": "dcat:Dataset",
      "odrl:hasPolicy": {
        "@id": "c2Vuc2l0aXZlLW9ubHktZGVm:YXNzZXQtMg==:MzhiYzZkNjctMDIyNi00OGJjLWFmNWYtZTQ2ZjAwYTQzOWI2",
        "@type": "odrl:Offer",
        "odrl:permission": [],
        "odrl:prohibition": [],
        "odrl:obligation": {
          "odrl:action": {
            "@id": "use"
          },
          "odrl:constraint": {
            "odrl:leftOperand": {
              "@id": "DataAccess.level"
            },
            "odrl:operator": {
              "@id": "odrl:eq"
            },
            "odrl:rightOperand": "sensitive"
          }
        }
      },
      "dcat:distribution": [
        {
          "@type": "dcat:Distribution",
          "dct:format": {
            "@id": "HttpData-PULL"
          },
          "dcat:accessService": {
            "@id": "a6c7f3a3-8340-41a7-8154-95c6b5585532",
            "@type": "dcat:DataService",
            "dcat:endpointDescription": "dspace:connector",
            "dcat:endpointUrl": "http://localhost:8192/api/dsp",
            "dct:terms": "dspace:connector",
            "dct:endpointUrl": "http://localhost:8192/api/dsp"
          }
        },
        {
          "@type": "dcat:Distribution",
          "dct:format": {
            "@id": "HttpData-PUSH"
          },
          "dcat:accessService": {
            "@id": "a6c7f3a3-8340-41a7-8154-95c6b5585532",
            "@type": "dcat:DataService",
            "dcat:endpointDescription": "dspace:connector",
            "dcat:endpointUrl": "http://localhost:8192/api/dsp",
            "dct:terms": "dspace:connector",
            "dct:endpointUrl": "http://localhost:8192/api/dsp"
          }
        }
      ],
      "description": "This asset requires Membership to view and SensitiveData credential to negotiate.",
      "id": "asset-2"
    }
  ],
  "dcat:distribution": [],
  "dcat:service": {
    "@id": "a6c7f3a3-8340-41a7-8154-95c6b5585532",
    "@type": "dcat:DataService",
    "dcat:endpointDescription": "dspace:connector",
    "dcat:endpointUrl": "http://localhost:8192/api/dsp",
    "dct:terms": "dspace:connector",
    "dct:endpointUrl": "http://localhost:8192/api/dsp"
  },
  "dspace:participantId": "did:web:localhost%3A7093",
  "participantId": "did:web:localhost%3A7093",
  "@context": {}
}

Catalogs are ephemeral objects, they are not persisted or cached on the provider side. Everytime a consumer participant makes a catalog request through DSP, the connector runtime has to evaluate the incoming request and build up the catalog specifically for that participant. The reason for this is that between two subsequent requests from the same participant, the contract definition or the claims or the participant could have changed.

The relevant component in EDC is the DatasetResolver, which resolves all contract definitions that are relevant to a participant filtering out those where the participant does not satisfy the access policy and collects all the assets therein.

In order to determine how an asset can be distributed, the resolver requires knowledge about the data planes that are available. It uses the Dataplane Signaling Protocol to query them and construct the list of Distributions for an asset.

For details about the FederatedCatalog, please refer to its documentation.

7 Transfer processes

A TransferProcess is a record of the data sharing procedure between a consumer and a provider. As they traverse through the system, they transition through several states (TransferProcessStates).

Once a contract is negotiated and an agreement is reached, the consumer connector may send a transfer initiate request to start the transfer. In the course of doing that, both parties may provision additional resources, for example deploying a temporary object store, where the provider should put the data. Similarly, the provider may need to take some preparatory steps, e.g. anonymizing the data before sending it out.

This is sometimes referred to as the provisioning phase. If no additional provisioning is needed, the transfer process simply transitions through the state with a NOOP.

Once that is done, the transfer begins in earnest. Data is transmitted according to the dataDestination, that was passed in the initiate-request.

Once the transmission has completed, the transfer process will transition to the COMPLETED state, or - if an error occurred - to the TERMINATED state.

The Management API provides several endpoints to manipulate data transfers.

Here is a diagram of the state machine applied to transfer processes on consumer side:

Transfer Process Consumer State Machine

Here is a diagram of the state machine applied to transfer processes on provider side:

Transfer Process Provider State Machine

A transfer process can be initiated from the consumer side by sending a TransferRequest to the connector Management API:

{
  "@context": {
    "@vocab": "https://w3id.org/edc/v0.0.1/ns/"
  },
  "@type": "https://w3id.org/edc/v0.0.1/ns/TransferRequest",
  "protocol": "dataspace-protocol-http",
  "counterPartyAddress": "http://provider-address",
  "contractId": "contract-id",
  "transferType": "transferType",
  "dataDestination": {
    "type": "data-destination-type"
  },
  "privateProperties": {
    "private-key": "private-value"
  },
  "callbackAddresses": [
    {
      "transactional": false,
      "uri": "http://callback/url",
      "events": [
        "contract.negotiation",
        "transfer.process"
      ],
      "authKey": "auth-key",
      "authCodeId": "auth-code-id"
    }
  ]
}

where:

7.1 Transfer and data flows types

The transfer type defines the channel (Distribution) for the data transfer and it depends on the capabilities of the data plane if it can be fulfilled. The transferType available for a data offering is available in the dct:format of the Distribution when inspecting the catalog response.

Each transfer type also characterizes the type of the flow, which can be either pull or push and it’s data can be either finite or non-finite

7.1.1 Consumer Pull

A pull transfer is when the consumer receives information (in the form of a DataAddress) on how to retrieve data from the Provider. Then it’s up to the consumer to use this information for pulling the data.

Consumer Pull

Provider and consumer agree to a contract (not displayed in the diagram)

  1. Consumer initiates the transfer process by sending a TransferRequestMessage
  2. The Provider Control Plane retrieves the DataAddress of the actual data source and creates aDataFlowStartMessage.
  3. The Provider Control Plane asks the selector which Data Plane instance can be used for this data transfer
  4. The Selector returns an eligible Data Plane instance (if any)
  5. Provider Control Plane sends the DataFlowStartMessage to the selected Data Plane instance through data plane signaling protocol.
  6. The Provider DataPlaneManager validates the incoming request and delegates to the DataPlaneAuthorizationService the generation of DataAddress, containing the information on location and authorization for fetching the data
  7. The Provider Data Plane acknowledges the Provider control plane and attach the DataAddress generated.
  8. The Provider Control Plane notifies the start of the transfer attaching the DataAddress in the TransferStartMessage.
  9. The Consumer Control plane receives the DataAddress and dispatch it accordingly to the configured runtime. Consumer can either decide to receive the DataAddress using the eventing system callbacks using the transfer.process.started type, or use the EDRs extensions for automatically store it on consumer control plane side.
  10. With the informations in the DataAddress such as the endpointUrl and the Authorization data can be fetched.
  11. The Provider Data plane validates and authenticates the incoming request and retrieves the source DataAddress.
  12. The he provider data plane proxies the validated request to the configured backend in the source DataAddress.

7.1.2 Provider Push

A push transfer is when the Provider data plane initiates sending data to the destination specified by the consumer.

Provider Push

Provider and consumer agree to a contract (not displayed in the diagram)

  1. The Consumer initiates the transfer process, i.e. sends TransferRequestMessage with a destination DataAddress
  2. The Provider Control Plane retrieves the DataAddress of the actual data source and creates a DataFlowStartMessage with both source and destination DataAddress.
  3. The Provider Control Plane asks the selector which Data Plane instance can be used for this data transfer
  4. The Selector returns an eligible Data Plane instance (if any)
  5. The Provider Control Plane sends the DataFlowStartMessage to the selected Data Plane instance through data plane signaling protocol.
  6. The Provider Data Plane validates the incoming request
  7. If request is valid, the Provider Data Plane returns acknowledgement
  8. The DataPlaneManager of the the Provider Data Plane processes the request: it creates a DataSource/DataSinkpair based on the source/destination data addresses
  9. The Provider Data Plane fetches data from the actual data source (see DataSource)
  10. The Provider Data Plane pushes data to the consumer services (see DataSink)

7.1.2 Finite and Non-Finite Data

The charaterization of the data applies to either push and pull transfers. Finite data transfers cause the transfer process to transitition to the state COMPLETED, once the transmission has finished. For example a transfer of a single file that is hosted and transferred into a cloud storage system.

Non-finite data means that once the transfer process request has been accepted by the provider the transfer process is in the STARTED state until it gets terminated by the consumer or the provider. Exampes of Non-finite data are streams or API endpoins.

On the provider side transfer processes can also be terminated by the policy monitor that periodically watches over the on going transfer and checks if the associated contract agreement still fulfills the contract policy.

7.2 About Data Destinations

A data destination is a description of where the consumer expects to find the data after the transfer completes. In a " provider-push" scenario this could be an object storage container, a directory on a file system, etc. In a “consumer-pull” scenario this would be a placeholder, that does not contain any information about the destination, as the provider “decides” which endpoint he makes the data available on.

A data address is a schemaless object, and the provider and the consumer need to have a common understanding of the required fields. For example, if the provider is supposed to put the data into a file share, the DataAddress object representing the data destination will likely contain the host URL, a path and possibly a file name. So both connectors need to be “aware” of that.

The actual data transfer is handled by a data plane through extensions ( called “sources” and " sinks"). Thus, the way to establish that “understanding” is to make sure that both parties have matching sources and sinks. That means, if a consumer asks to put the data in a file share, the provider must have the appropriate data plane extensions to be able to perform that transfer.

If the provider connector does not have the appropriate extensions loaded at runtime, the transfer process will fail.

7.3 Transfer process callbacks

In order to get timely updates about status changes of a transfer process, we could simply poll the management API by firing a GET /v*/transferprocesses/{tp-id}/state request every X amount of time. That will not only put unnecessary load on the connector, you may also run into rate-limiting situations, if the connector is behind a load balancer of some sort. Thus, we recommend using event callbacks.

Callbacks must be specified when requesting to initiate the transfer:

{
  // ...
  "callbackAddresses": [
    {
      "transactional": false,
      "uri": "http://callback/url",
      "events": [
        "transfer.process"
      ],
      "authKey": "auth-key",
      "authCodeId": "auth-code-id"
    }
  ]
  //...
}

Currently, we support the following events:

  • transfer.process.deprovisioned
  • transfer.process.completed
  • transfer.process.deprovisioningRequested
  • transfer.process.initiated
  • transfer.process.provisioned
  • transfer.process.provisioning
  • transfer.process.requested
  • transfer.process.started
  • transfer.process.terminated

The connector’s event dispatcher will send invoke the webhook specified in the uri field passing the event payload as JSON object.

More info about events and callbacks can be found here.

8 Endpoint Data References

9 Querying with QuerySpec and Criterion

Most of the entities can be queried with the QuerySpec object, which is a generic way of expressing limit, offset, sort and filters when querying a collection of objects managed by the EDC stores.

Here’s an example of how a QuerySpec object might look like when querying for Assets via management APIs:

{
  "@context": {
    "edc": "https://w3id.org/edc/v0.0.1/ns/"
  },
  "@type": "QuerySpec",
  "limit": 1,
  "offset": 1,
  "sortField": "createdAt",
  "sortOrder": "DESC",
  "filterExpression": [
    {
      "operandLeft": "https://w3id.org/edc/v0.0.1/ns/description",
      "operator": "=",
      "operandRight": "This asset"
    }
  ]
}

which filters by the description custom property being equals to This asset. The query also paginates the result with limit and p set to 1. Additionally a sorting strategy is in place by createdAt property in descending order ( the default is ASC)

Note: Since custom properties are persisted in their expanded form, we have to use the expanded form also when querying.

The filterExpression property is a list of Criterion, which expresses a single filtering condition based on:

  • operandLeft: the property to filter on
  • operator: the operator to apply e.g. =
  • operandRight: the value of the filtering

The supported operators are:

  • Equal: =
  • Not equal: !=
  • In: in
  • Like: like
  • Ilike: ilike (same as like but ignoring case sensitive)
  • Contains: contains

Note: multiple filtering expressions are always logically conjoined using an “AND” operation.

The properties that can be expressed in the operandLeft of a Criterion depend on the shape of the entity that we are want to query.

Note: nested properties are also supported using the dot notation.

QuerySpec can also be used when doing the catalog request using the querySpec property in the catalog request payload for filtering the datasets:

{
  "@context": {
    "@vocab": "https://w3id.org/edc/v0.0.1/ns/"
  },
  "counterPartyAddress": "http://provider/api/dsp",
  "protocol": "dataspace-protocol-http",
  "counterPartyId": "providerId",
  "querySpec": {
    "filterExpression": [
      {
        "operandLeft": "https://w3id.org/edc/v0.0.1/ns/description",
        "operator": "=",
        "operandRight": "This asset"
      }
    ]
  }
}

Entities are backed by stores for doing CRUD operations. For each entity there is an associated store interface (SPI). Most of the stores SPI have a query like method which takes a QuerySpec type as input and returns the matched entities in a collection. Indivitual implementations are then responsible for translating the QuerySpec to a proper fetching strategy.

The description on how the translation and mapping works will be explained in each implementation. Currently EDC support out of the box:

  • In-memory stores (default implementation).
  • SQL stores provied as extensions for each store, mostly tailored for and tested with PostgreSQL.

For guaranteeing the highest compatibility between store implementations, a base tests suite is provided for each store that each technology implementors need to fulfill in order to have a minimum usable store implementation.