daml/navigator/backend/README.md
Stefano Baghino f8bebffb4e Expose signatories and observers throughout the platform (#1814)
* Add signatories and observers to Ledger API

Cascade fixes deriving from compilation and test errors.

* Add signatories and observers to Java bindings and codegen

* Add signatories and observers to Scala bindings and codegen

* Add signatories and observers to navigator

* Add signatories and observers to the Haskell bindings

* Add integration test

* Comply with formatting standard

* Add release notes

* Apply edits from CI feedback and self-review

* Fix wrong input on test case for observers sorting

Co-Authored-By: Robert Autenrieth <31539813+rautenrieth-da@users.noreply.github.com>

* Read JSON as strings from DB, parse them as parties

Address https://github.com/digital-asset/daml/pull/1814#discussion_r296405252

* Change parties rendering and observer display condition

Address https://github.com/digital-asset/daml/pull/1814#discussion_r296390104

* Fix sorting tests, implement sorting by agreement

Address https://github.com/digital-asset/daml/pull/1814#discussion_r296397390

FIXME notes have been added for suspicious test cases

* Prevent any mutation of CreatedEvent arguments

Address https://github.com/digital-asset/daml/pull/1814#discussion_r296572639

* Comply with formatting standard

* Stakeholders are filtered up to the witnesses

Addresses https://github.com/digital-asset/daml/pull/1814#discussion_r297102521
2019-06-26 14:02:59 +00:00

8.8 KiB

Navigator Backend

The Navigator backend is a Scala application providing

  • a web server that exposes a GraphQL API with some predefined endpoints, such as returning all visible contracts.
  • a platform client that reacts to events happening in the platform
  • a simple sign-in API

Usage

The Navigator backend is written in Scala, making heavy use of Akka and Sangria (for GraphQL). It uses the Scala Build Tool.

We can build and run a basic the backend using the following commands:

# Build a distribution archive ("fat jar"):
bazel build //navigator/backend:navigator-binary_deploy.jar

# Run without arguments to show usage:
java -jar bazel-bin/navigator/backend/navigator-binary_deploy.jar --help

# Create a dummy configuration file
cat << EOF > navigator.conf
users {
  OPERATOR: {
    party=OPERATOR
    password=operator_password
  }
}
EOF

# Start the web server
java -jar bazel-bin/navigator/backend/navigator-binary_deploy.jar server -c navigator.conf

If you start the server and the configuration file doesn't exist, the server will write a configuration template and quit. You can then edit the configuration file and start again the server.

Custom UI backends

Extend the UIBackend base class

The base class for UI backends is the UIBackend, a Scala class that implements a ledger indexer and a web-server which exposes a number of default GraphQL endpoints and a simple Session API. It also allows has hooks for adding custom GraphQL endpoints.

To create a basic UI backend, you can extend UIBackend without any custom additions:

object DefaultUIBackend extends UIBackend {
  override def customEndpoints: Set[CustomEndpoint[_]] = Set()
  override val defaultConfigFile: Path = Paths.get("my-app.conf")
  override def applicationInfo: ApplicationInfo = ApplicationInfo(...)
  override def banner: Option[String] = Some(...)
}

You can then run your app with this as the main class.

Custom endpoints

A custom endpoint is an extra endpoint for a UI backend that binds a name to some data that can be represented in GraphQL and a function to calculate that data. For example:

object ExampleUIBackend extends UIBackend {

  override def customEndpoints: Set[CustomEndpoint[_]] = Set(contractsIdEndpoint)

  final case class TemplateId(id: String)
  final case class ContractProjection(id: String, template: TemplateId, argument: RecordArgument)

  /** Endpoing for the contract ids */
  private val contractsIdEndpoint = new CustomEndpoint[ContractProjection] {
    /** The endpoint to be used as GraphQL top-level for the data served by this */
    override def endpointName: String = "contract_id"

    /** For each contract in the ledger, extract contract id and template id */
    override def calculate(ledger: Ledger): Seq[ContractProjection] =
      ledger.allContracts().map(contract =>
        ContractProjection(contract.idString, TemplateId(contract.template.idString), contract.argument))
  }
}

This adds an endpoint called contract_id that returns a list of contract projections that include the contract ID, the template ID, and the contract arguments. You would then pass this custom endpoint to a UI backend subclass:

There are two components of a CustomEndpoint:

  • The endpointName is the name used to bind this custom endpoint to the root of the GraphQL Schema. Note that each endpoint must have a unique endpointName and that default endpointNames -- contracts or templates for example -- cannot be used in custom endpoints either.

  • The calculate function is used to extract the data to serve from the Ledger. This function must create a sequence of instances of a scala case class.

The endpoint contract_id becomes available in the root of the graphql schema:

type ContractProjection {
  id: String!
  template: TemplateId!
  argument: Value!
}

type TemplateId {
  id: String!
}

type Query {
  contract_id: [ContractProjection!]!
}

The schema is automatically generated by the UI backend by following the structure of a "row" case class. For instance TemplateId is a nested structure inside ContractProjection in the Scala code and for this reason it will be a nested structure in the GraphQL Schema too.

Automatic generation of GraphQL Schemas for custom endpoints

The code for the GraphQL endpoint is automatically generated at compile time by the UI backend. This means that there are some limitations in what can be exposed as result of the calculate function. Specifically:

  • The name of the "row" model case class must be a unique name. Don't use Contract or Template or any other name already used in the GraphQL Schema. For example, you can't rename ContractProjection to Contract.

  • The main "row" class must be modelled as a Scala case class and it must contain fields that are either a case class or Scala primitives. If something is not supported, you won't be able to compile your code.

If some type is not supported by the generator, it is possible to add support for it by providing either an instance of the typeclass GraphQLLeaf for simple types, e.g. String or Int, or an instance of the typeclass GraphQLObject for complex types which have fields.

Common tasks

How do I query the Backend without a frontend?

The UI backend exposes a GraphiQL page to let the developer run queries interactively. You can find it at the address <address_of_the_navigator_backend>/graphql. Note the "Docs" link in the upper right corner and also that you have to be logged in order to run queries. You can either use the frontend in the same browser to login or you can login manually via command line as follows:

  • Send a JSON POST request with the userId and the password and make note of the cookie:

    > curl -H "Content-Type: application/json" -d'{ "userId":"PARTY", "password":"password" }' localhost:4000/api/session/ -i
    HTTP/1.1 200 OK
    Set-Cookie: session-id=8b4601d4-7113-407b-9b81-7fd5b213a96b; Path=/
    Server: akka-http/10.0.4
    Date: Tue, 13 Jun 2017 16:47:38 GMT
    Content-Type: application/json
    Content-Length: 92
    
    {"type":"session","user":{"id":"BANK1","role":"bank","party":"BANK1","canAdvanceTime":true}}
    
  • Open the GraphiQL page in a browser (localhost:4000/graphql for example).

  • Write javascript:document.cookie="session-id=<the_id_recceived>" in the location bar.

  • You can now use the GraphiQL to query your data.

We plan to add an option to run the server in unauthenticated mode so that these steps become unnecessary.

Session API

In addition to the GraphQL endpoint exposing data, the user needs to act as some party. This is to know which party's "view" on the ledger to expose. The backend supplies the frontend with a list of available users. Alternatively, if the backend is started with the --require-password flag, it will not supply a list of users and instead ask the frontend to present the user with a standard username/password form. The chosen user is set in a cookie and therefore persists across reloads.

type UserId = string;
type Party = string;
type Role = string;
type User = {
  id: UserId;
  party: Party;
  canAdvanceTime: boolean;
  role?: Role;
};

type Status = Session | SignIn;
type Session = { type: 'session'; user: User; }
type SignIn = {
  type: 'sign-in';
  method: SignInMethod;
  error?: 'invalid-credentials';
}

type SignInMethod = SignInPassword | SignInSelect
type SignInPassword = { type: 'password' }
type SignInSelect = { type: 'select', users: UserId[] }
# Get current session or sign in
GET /session/ => Status

# Sign in
POST /session/ -d'{ userId: UserId, password?: String }' => Status

# Sign out
DELETE /session/ => SignIn

Architecture notes

This section contains short notes for anyone that wishes to make changes to the UI backend library.

  • The backend defines its own internal representation of ledger objects
  • The backend communicates with the ledger via a JSON-based "ledger API"
  • The backend communicates with the frontend via a JSON-based "frontend API"
  • These are the relevant source files:
    • model/Model: Defines the internal representations of ledger objects
    • model/Util: (SDAML package) -> (internal model)
    • graphql/JsonType: (internal model) <-> (frontend API JSON format)
    • store/platform/PlatformSubscriber: (ledger API JSON format) <-> (internal model)