17 KiB
Brief introduction to the node services
This document is intended as a very brief introduction to the current service components inside the node. Whilst not at all exhaustive it is hoped that this will give some context when writing applications and code that use these services, or which are operated upon by the internal components of Corda.
Services within the node
The node services represent the various sub functions of the Corda node. Some are directly accessible to contracts and flows through the ServiceHub, whilst others are the framework internals used to host the node functions. Any public service interfaces are defined in the :core gradle project in the src/main/kotlin/net/corda/core/node/services folder. The ServiceHub interface exposes functionality suitable for flows. The implementation code for all standard services lives in the gradle :node project under the src/main/kotlin/net/corda/node/services folder. The src/main/kotlin/net/corda/node/services/api folder contains declarations for internal only services and for interoperation between services.
All the services are constructed in the AbstractNode start method (and the extension in Node). They may also register a shutdown handler during initialisation, which will be called in reverse order to the start registration sequence when the Node.stop is called.
As well as the standard services trusted CorDapp plugins may register custom services. These plugin services are passed a reference to the PluginServiceHub which allows some more powerful functions e.g. starting flows.
For unit testing a number of non-persistent, memory only services are defined in the :node and :test-utils projects. The :test-utils project also provides an in-memory networking simulation to allow unit testing of flows and service functions.
The roles of the individual services are described below.
Key management and identity services
InMemoryIdentityService
The InMemoryIdentityService implements the IdentityService interface and provides a store of remote mappings between CompositeKey and remote Parties. It is automatically populated from the NetworkMapCache updates and is used when translating CompositeKey exposed in transactions into fully populated Party identities. This service is also used in the default JSON mapping of parties in the web server, thus allowing the party names to be used to refer to other node legal identities. In the future the Identity service will be made persistent and extended to allow anonymised session keys to be used in flows where the well-known CompositeKey of nodes need to be hidden to non-involved parties.
PersistentKeyManagementService and E2ETestKeyManagementService
Typical usage of these services is to locate an appropriate PrivateKey to complete and sign a verified transaction as part of a flow. The normal node legal identifier keys are typically accessed via helper extension methods on the ServiceHub, but these ultimately fetch the keys from the KeyManagementService. The KeyManagementService interface also allows other keys to be generated if anonymous keys are needed in a flow. Note that this interface works at the level of individual PublicKey/PrivateKey pairs, but the signing authority will be represented by a CompositeKey on the NodeInfo to allow key clustering and threshold schemes.
The PersistentKeyManagementService is a persistent implementation of the KeyManagementService interface that records the key pairs to a key-value storage table in the database. E2ETestKeyManagementService is a simple implementation of the KeyManagementService that is used to track our KeyPairs for use in unit testing when no database is available.
Messaging and network management services
ArtemisMessagingServer
The ArtemisMessagingServer service is run internally by the Corda node to host the ArtemisMQ messaging broker that is used for reliable node communications. Although the node can be configured to disable this and connect to a remote broker by setting the messagingServerAddress configuration to be the remote broker address. (The MockNode used during testing does not use this service, and has a simplified in-memory network layer instead.) This service is not exposed to any CorDapp code as it is an entirely internal infrastructural component. However, the developer may need to be aware of this component, because the ArtemisMessagingServer is responsible for configuring the network ports (based upon settings in node.conf) and the service configures the security settings of the ArtemisMQ middleware and acts to form bridges between node mailbox queues based upon connection details advertised by the NetworkMapService. The ArtemisMQ broker is configured to use TLS1.2 with a custom TrustStore containing a Corda root certificate and a KeyStore with a certificate and key signed by a chain back to this root certificate. These keystores typically reside in the certificates sub folder of the node workspace. For the nodes to be able to connect to each other it is essential that the entire set of nodes are able to authenticate against each other and thus typically that they share a common root certificate. Also note that the address configuration defined for the server is the basis for the address advertised in the NetworkMapService and thus must be externally connectable by all nodes in the network.
NodeMessagingClient
The NodeMessagingClient is the implementation of the MessagingService interface operating across the ArtemisMQ middleware layer. It typically connects to the local ArtemisMQ hosted within the ArtemisMessagingServer service. However, the messagingServerAddress configuration can be set to a remote broker address if required. The responsibilities of this service include managing the node's persistent mailbox, sending messages to remote peer nodes, acknowledging properly consumed messages and deduplicating any resent messages. The service also handles the incoming requests from new RPC client sessions and hands them to the CordaRPCOpsImpl to carry out the requests.
InMemoryNetworkMapCache
The InMemoryNetworkMapCache implements the NetworkMapCache interface and is responsible for tracking the identities and advertised services of authorised nodes provided by the remote NetworkMapService. Typical use is to search for nodes hosting specific advertised services e.g. a Notary service, or an Oracle service. Also, this service allows mapping of friendly names, or Party identities to the full NodeInfo which is used in the StateMachineManager to convert between the CompositeKey, or Party based addressing used in the flows/contracts and the physical host and port information required for the physical ArtemisMQ messaging layer.
PersistentNetworkMapService and NetworkMapService
The NetworkMapService is a node internal component responsible for managing and communicating the directory of authenticated registered nodes and advertised services in the Corda network. Only a single node in the network (in future this will be a clustered service) should host the NetworkMapService implementation. All other Corda nodes initiate their remote connection to the NetworkMapService early in the start-up sequence and wait to synchronise their local NetworkMapCache before activating any flows. For the PersistentNetworkMapService registered NodeInfo data is persisted and will include nodes that are not currently active. The networking layer will persist any messages directed at such inactive nodes with the expectation that they will be delivered eventually, or else that the source flow will be terminated by admin intervention. An InMemoryNetworkMapService is also available for unit tests without a database.
The NetworkMapService should not be used by any flows, or contracts. Instead they should access the NetworkMapCache service to access this data.
Storage and persistence related services
StorageServiceImpl
The StorageServiceImpl service simply hold references to the various persistence related services and provides a single grouped interface on the ServiceHub.
DBCheckpointStorage
The DBCheckpointStorage service is used from within the StateMachineManager code to persist the progress of flows. Thus ensuring that if the program terminates the flow can be restarted from the same point and complete the flow. This service should not be used by any CorDapp components.
DBTransactionMappingStorage and InMemoryStateMachineRecordedTransactionMappingStorage
The DBTransactionMappingStorage is used within the StateMachineManager code to relate transactions and flows. This relationship is exposed in the eventing interface to the RPC clients, thus allowing them to track the end result of a flow and map to the actual transactions/states completed. Otherwise this service is unlikely to be accessed by any CorDapps. The InMemoryStateMachineRecordedTransactionMappingStorage service is available as a non-persistent implementation for unit tests with no database.
DBTransactionStorage
The DBTransactionStorage service is a persistent implementation of the TransactionStorage interface and allows flows read-only access to full transactions, plus transaction level event callbacks. Storage of new transactions must be made via the recordTransactions method on the ServiceHub, not via a direct call to this service, so that the various event notifications can occur.
NodeAttachmentService
The NodeAttachmentService provides an implementation of the AttachmentStorage interface exposed on the ServiceHub allowing transactions to add documents, copies of the contract code and binary data to transactions. The data is persisted to the local file system inside the attachments subfolder of the node workspace. The service is also interfaced to by the web server, which allows files to be uploaded via an HTTP post request.
Flow framework and event scheduling services
StateMachineManager
The StateMachineManager is the service that runs the active flows of the node whether initiated by an RPC client, the web interface, a scheduled state activity, or triggered by receipt of a message from another node. The StateMachineManager wraps the flow code (extensions of the FlowLogic class) inside an instance of the FlowStateMachineImpl class, which is a Quasar Fiber. This allows the StateMachineManager to suspend flows at all key lifecycle points and persist their serialized state to the database via the DBCheckpointStorage service. This process uses the facilities of the Quasar Fibers library to manage this process and hence the requirement for the node to run the Quasar java instrumentation agent in its JVM.
In operation the StateMachineManager is typically running an active flow on its server thread until it encounters a blocking, or externally visible operation, such as sending a message, waiting for a message, or initiating a subFlow. The fiber is then suspended and its stack frames serialized to the database, thus ensuring that if the node is stopped, or crashes at this point the flow will restart with exactly the same action again. To further ensure consistency, every event which resumes a flow opens a database transaction, which is committed during this suspension process ensuring that the database modifications e.g. state commits stay in sync with the mutating changes of the flow. Having recorded the fiber state the StateMachineManager then carries out the network actions as required (internally one flow message exchanged may actually involve several physical session messages to authenticate and invoke registered flows on the remote nodes). The flow will stay suspended until the required message is returned and the scheduler will resume processing of other activated flows. On receipt of the expected response message from the network layer the StateMachineManager locates the appropriate flow, resuming it immediately after the blocking step with the received message. Thus from the perspective of the flow the code executes as a simple linear progression of processing, even if there were node restarts and possibly message resends (the messaging layer deduplicates messages based on an id that is part of the checkpoint).
The StateMachineManager service is not directly exposed to the flows, or contracts themselves.
NodeSchedulerService
The NodeSchedulerService implements the SchedulerService interface and monitors the Vault updates to track any new states that implement the SchedulableState interface and require automatic scheduled flow initiation. At the scheduled due time the NodeSchedulerService will create a new flow instance passing it a reference to the state that triggered the event. The flow can then begin whatever action is required. Note that the scheduled activity occurs in all nodes holding the state in their Vault, it may therefore be required for the flow to exit early if the current node is not the intended initiator.
Notary flow implementation services
PersistentUniquenessProvider, InMemoryUniquenessProvider and RaftUniquenessProvider
These variants of UniquenessProvider service are used by the notary flows to track consumed states and thus reject double-spend scenarios. The InMemoryUniquenessProvider is for unit testing only, the default being the PersistentUniquenessProvider which records the changes to the DB. When the Raft based notary is active the states are tracked by the whole cluster using a RaftUniquenessProvider. Outside of the notary flows themselves this service should not be accessed by any CorDapp components.
NotaryService (SimpleNotaryService, ValidatingNotaryService, RaftValidatingNotaryService)
The NotaryService is an abstract base class for the various concrete implementations of the Notary server flow. By default, a node does not run any NotaryService server component. However, the appropriate implementation service is automatically started if the relevant ServiceType id is included in the node's extraAdvertisedServiceIds configuration property. The node will then advertise itself as a Notary via the NetworkMapService and may then participate in controlling state uniqueness when contacted by nodes using the NotaryFlow.Client subFlow. The SimpleNotaryService only offers protection against double spend, but does no further verification. The ValidatingNotaryService checks that proposed transactions are correctly signed by all keys listed in the commands and runs the contract verify to ensure that the rules of the state transition are being followed. The RaftValidatingNotaryService further extends the flow to operate against a cluster of nodes running shared consensus state across the RAFT protocol (note this requires the additional configuration of the notaryClusterAddresses property).
Vault related services
NodeVaultService
The NodeVaultService implements the VaultService interface to allow access to the node's own set of unconsumed states. The service does this by tracking update notifications from the TransactionStorage service and processing relevant updates to delete consumed states and insert new states. The resulting update is then persisted to the database. The VaultService then exposes query and event notification APIs to flows and CorDapp plugins to allow them to respond to updates, or query for states meeting various conditions to begin the formation of new transactions consuming them. The equivalent services are also forwarded to RPC clients, so that they may show updating views of states held by the node.
NodeSchemaService and HibernateObserver
The HibernateObserver runs within the node framework and listens for vault state updates, the HibernateObserver then uses the mapping services of the NodeSchemaService to record the states in auxiliary database tables. This allows Corda state updates to be exposed to external legacy systems by insertion of unpacked data into existing tables. To enable these features the contract state must implement the QueryableState interface to define the mappings.