State Persistence
Corda offers developers the option to expose all or some parts of a contract state to an Object Relational Mapping (ORM) tool to be persisted in a Relational Database Management System (RDBMS).
This assists vault development and allows the persistence of state data to a custom database table. Persisted states held in the vault are indexed to execute queries. This allows relational joins between Corda tables and the organization’s existing data.
The Object Relational Mapping is specified using Java Persistence API (JPA) annotations. This mapping is persisted to the database as a table row (a single, implicitly structured data item) by the node automatically every time a state is recorded in the node’s local vault as part of a transaction.
Schemas
All ContractState
s may implement the QueryableState
interface. The interface inserts the ContractState
into a custom table in the node’s database, accessible using SQL.
/**
* A contract state that may be mapped to database schemas configured for this node to support querying for,
* or filtering of, states.
*/
@KeepForDJVM
interface QueryableState : ContractState {
/**
* Enumerate the schemas this state can export representations of itself as.
*/
fun supportedSchemas(): Iterable<MappedSchema>
/**
* Export a representation for the given schema.
*/
fun generateMappedObject(schema: MappedSchema): PersistentState
}
The QueryableState
interface requires the state to enumerate the different relational schemas it supports, for
instance in situations where the schema has evolved. Each relational schema is represented as a MappedSchema
object returned by the state’s supportedSchemas
method.
Nodes have an internal SchemaService
which decides what data to persist by selecting the MappedSchema
to use.
Once a MappedSchema
is selected, the SchemaService
will delegate to the QueryableState
to generate a corresponding
representation (mapped object) via the generateMappedObject
method, the output of which is then passed to the ORM.
/**
* A configuration and customisation point for Object Relational Mapping of contract state objects.
*/
interface SchemaService {
/**
* All available schemas in this node
*/
val schemas: Set<MappedSchema>
/**
* Given a state, select schemas to map it to that are supported by [generateMappedObject] and that are configured
* for this node.
*/
fun selectSchemas(state: ContractState): Iterable<MappedSchema>
/**
* Map a state to a [PersistentState] for the given schema, either via direct support from the state
* or via custom logic in this service.
*/
fun generateMappedObject(state: ContractState, schema: MappedSchema): PersistentState
}
/**
* A database schema that might be configured for this node. As well as a name and version for identifying the schema,
* also list the classes that may be used in the generated object graph in order to configure the ORM tool.
*
* @param schemaFamily A class to fully qualify the name of a schema family (i.e. excludes version)
* @param version The version number of this instance within the family.
* @param mappedTypes The JPA entity classes that the ORM layer needs to be configure with for this schema.
*/
@KeepForDJVM
open class MappedSchema(schemaFamily: Class<*>,
val version: Int,
val mappedTypes: Iterable<Class<*>>) {
val name: String = schemaFamily.name
/**
* Optional classpath resource containing the database changes for the [mappedTypes]
*/
open val migrationResource: String? = null
override fun toString(): String = "${this.javaClass.simpleName}(name=$name, version=$version)"
override fun equals(other: Any?): Boolean {
if (this === other) return true
if (javaClass != other?.javaClass) return false
other as MappedSchema
if (version != other.version) return false
if (mappedTypes != other.mappedTypes) return false
if (name != other.name) return false
return true
}
override fun hashCode(): Int {
var result = version
result = 31 * result + mappedTypes.hashCode()
result = 31 * result + name.hashCode()
return result
}
}
With this framework, the relational view of ledger states can evolve in a controlled fashion in lock-step with internal systems or other integration points, and is not dependent on changes to the contract code.
Multiple contract state implementations might provide mappings within a single schema. For example, an Interest Rate Swap contract and an Equity OTC Option contract might both provide a mapping to a derivative contract within the same schema. The schemas should typically not be part of the contract itself and should exist independently to encourage reuse of a common set within a particular business area or CorDapp.
MappedSchema
or migrating its data. At startup, nodes log such violations as warnings stating that there’s a cross-reference between MappedSchema
’s.
The detailed messages incorporate information about what schemas, entities and fields are involved.MappedSchema
offers a family name, which is disambiguated using Java-package-style name-spacing derived from the
class name of a schema family class that is constant across versions. This allows the SchemaService
to select a
preferred version of a schema.The SchemaService
is also responsible for the SchemaOptions
you can configure for a particular
MappedSchema
. These allow the configuration of database schemas or table name prefixes to avoid clashes with
other MappedSchema
.
Custom schema registration
Custom contract schemas register automatically when CorDapps start up. The node bootstrap process scans for states that implement the QueryableState
interface. The MappedSchema
specifies which tables to create, as identified by each state’s supportedSchemas
method.
For testing purposes you must manually register the packages containing custom schemas:
- Tests using
MockNetwork
andMockNode
must explicitly register packages using the cordappPackages parameter ofMockNetwork
. - Tests using
MockServices
must explicitly register packages using the cordappPackages parameter of theMockServices
makeTestDatabaseAndMockServices() helper method.
Object relational mapping
To facilitate the object relational mapping (ORM), the persisted representation of a QueryableState
should be an instance of a PersistentState
subclass,constructed either by the state itself or a plugin to the SchemaService
. This allows the ORM layer to always
associate a StateRef
with a persisted representation of a ContractState
and allows joining with the set of
unconsumed states in the vault.
The PersistentState
subclass should be marked up as a JPA 2.1 Entity with a defined table name and having
properties (in Kotlin, getters/setters in Java) annotated to map to the appropriate columns and SQL types. Additional
entities can be included to model these properties where they are more complex. The mapping does not need to be flat - for example, for collections. The MappedSchema
constructor accepts a list of all JPA entity classes for that schema in
the MappedTypes
parameter. It must provide this list in order to initialise the ORM layer.
Several examples of entities and mappings are provided in the codebase, including Cash.State
and
CommercialPaper.State
. For example, the first version of the cash schema:
package net.corda.finance.schemas
import net.corda.core.identity.AbstractParty
import net.corda.core.schemas.MappedSchema
import net.corda.core.schemas.PersistentState
import net.corda.core.serialization.CordaSerializable
import net.corda.core.utilities.MAX_HASH_HEX_SIZE
import net.corda.core.contracts.MAX_ISSUER_REF_SIZE
import org.hibernate.annotations.Type
import javax.persistence.*
/**
* An object used to fully qualify the [CashSchema] family name (i.e. independent of version).
*/
object CashSchema
/**
* First version of a cash contract ORM schema that maps all fields of the [Cash] contract state as it stood
* at the time of writing.
*/
@Suppress("MagicNumber") // SQL column length
@CordaSerializable
object CashSchemaV1 : MappedSchema(schemaFamily = CashSchema.javaClass, version = 1, mappedTypes = listOf(PersistentCashState::class.java)) {
override val migrationResource = "cash.changelog-master"
@Entity
@Table(name = "contract_cash_states", indexes = [Index(name = "ccy_code_idx", columnList = "ccy_code"), Index(name = "pennies_idx", columnList = "pennies")])
class PersistentCashState(
/** X500Name of owner party **/
@Column(name = "owner_name", nullable = true)
var owner: AbstractParty?,
@Column(name = "pennies", nullable = false)
var pennies: Long,
@Column(name = "ccy_code", length = 3, nullable = false)
var currency: String,
@Column(name = "issuer_key_hash", length = MAX_HASH_HEX_SIZE, nullable = false)
var issuerPartyHash: String,
@Column(name = "issuer_ref", length = MAX_ISSUER_REF_SIZE, nullable = false)
@Type(type = "corda-wrapper-binary")
var issuerRef: ByteArray
) : PersistentState()
}
Persisting Hierarchical Data
You may want to persist hierarchical relationships within state data using multiple database tables. To facilitate this, you must implement all queries making use of hierarchical relations as native SQL.
Example schemas implementing hierarchical relationships have been implemented below.
For one-to-one scenarios, use the following:
// Parent schema implemented as usual
class PersistentParent(
@Column(name = "linear_id")
var linearId: UUID
) : PersistentState() {
constructor() : this( UUID.randomUUID() )
}
// Child has a reference to its parent
class PersistentChild(
@Column(name = "linear_id")
var linearId: UUID
@Column=(name = "parent_linear_id")
var parentLinearId: UUID
) : PersistentState() {
constructor() : this( UUID.randomUUID(), UUID.randomUUID() )
}
For one-to-many scenarios, use the following:
// Parent schema implemented as usual
class PersistentParent(
@Column(name = "linear_id")
var linearId: UUID
) : PersistentState() {
constructor() : this( UUID.randomUUID() )
}
// Child schema implemented as usual
class PersistentChild(
@Column(name = "linear_id")
var linearId: UUID
) : PersistentState() {
constructor() : this( UUID.randomUUID())
}
// ParentChildThrough table schema
class PersistentThroughTable(
@Column(name = "parent_linear_id")
var parentLinearId: UUID
@Column(name="child_linear_id")
var childLinearId: UUID
) : PersistentState() {
constructor() : this( UUID.randomUUID(), UUID.randomUUID() )
}
Identity mapping
Schema entity attributes defined by identity types (AbstractParty
, Party
, AnonymousParty
) are automatically
processed to ensure only the X500Name
of the identity is persisted where an identity is well known. Otherwise a null
value is stored in the associated column. To preserve privacy, identity keys are never persisted. Developers should use
the IdentityService
to resolve keys from well-known X500 identity names.
JDBC session
Apps can interact directly with the underlying node’s database using a standard JDBC connection (session) as described by the Java SQL Connection API
Use the ServiceHub
jdbcSession
function to get a JDBC connection:
val nativeQuery = "SELECT v.transaction_id, v.output_index FROM vault_states v WHERE v.state_status = 0"
database.transaction {
val jdbcSession = services.jdbcSession()
val prepStatement = jdbcSession.prepareStatement(nativeQuery)
val rs = prepStatement.executeQuery()
}
You can use JDBC sessions in flows and services. See Writing flows.
The following example illustrates the creation of a custom Corda service using a jdbcSession
:
object CustomVaultQuery {
@CordaService
class Service(val services: AppServiceHub) : SingletonSerializeAsToken() {
private companion object {
private val log = contextLogger()
}
fun rebalanceCurrencyReserves(): List<Amount<Currency>> {
val nativeQuery = """
select
cashschema.ccy_code,
sum(cashschema.pennies)
from
vault_states vaultschema
join
contract_cash_states cashschema
where
vaultschema.output_index=cashschema.output_index
and vaultschema.transaction_id=cashschema.transaction_id
and vaultschema.state_status=0
group by
cashschema.ccy_code
order by
sum(cashschema.pennies) desc
"""
log.info("SQL to execute: $nativeQuery")
val session = services.jdbcSession()
return session.prepareStatement(nativeQuery).use { prepStatement ->
prepStatement.executeQuery().use { rs ->
val topUpLimits: MutableList<Amount<Currency>> = mutableListOf()
while (rs.next()) {
val currencyStr = rs.getString(1)
val amount = rs.getLong(2)
log.info("$currencyStr : $amount")
topUpLimits.add(Amount(amount, Currency.getInstance(currencyStr)))
}
topUpLimits
}
}
}
}
}
which is then referenced within a custom flow:
@Suspendable
@Throws(CashException::class)
override fun call(): List<SignedTransaction> {
progressTracker.currentStep = AWAITING_REQUEST
val topupRequest = otherPartySession.receive<TopupRequest>().unwrap {
it
}
val customVaultQueryService = serviceHub.cordaService(CustomVaultQuery.Service::class.java)
val reserveLimits = customVaultQueryService.rebalanceCurrencyReserves()
val txns: List<SignedTransaction> = reserveLimits.map { amount ->
// request asset issue
logger.info("Requesting currency issue $amount")
val txn = issueCashTo(amount, topupRequest.issueToParty, topupRequest.issuerPartyRef, topupRequest.notaryParty)
progressTracker.currentStep = SENDING_TOP_UP_ISSUE_REQUEST
return@map txn.stx
}
otherPartySession.send(txns)
return txns
}
For examples of @CordaService
implementation testing, see the Oracle example.
Restricted control of connections
Corda restricts the functions available by the Connection
returned by jdbcSession
. This is to prevent a flow’s underlying database transaction from being tampered with, which could cause errors within the flow.
Calling jdbcSession
returns a RestrictedConnection
which prevents calls to the following functions:
abort(executor: Executor?)
clearWarnings()
close()
commit()
setSavepoint() methods
releaseSavepoint(savepoint: Savepoint?)
rollback() methods
setCatalog(catalog : String?)
setTransactionIsolation(level: Int)
setTypeMap(map: MutableMap<String, Class<*>>?)
setHoldability(holdability: Int)
setSchema(schema: String?)
setNetworkTimeout(executor: Executor?, milliseconds: Int)
setAutoCommit(autoCommit: Boolean)
setReadOnly(readOnly: Boolean)
JPA Support
In addition to jdbcSession
, ServiceHub
also exposes the Java Persistence API to flows via the withEntityManager
method. This method can be used to persist and query entities which inherit from MappedSchema
. This is particularly
useful if you need to maintain off-ledger data in conjunction with on-ledger state data.
mappedType
as part of a MappedSchema
. If you do not, no corresponding table will be created. See the examples below.The code snippet below defines a PersistentFoo
type inside FooSchemaV1
. Note that PersistentFoo
is added to
a list of mapped types which is passed to MappedSchema
. This is exactly how state schemas are defined, except that
the entity in this case should not subclass PersistentState
as it is not a state object:
public class FooSchema {}
public class FooSchemaV1 extends MappedSchema {
FooSchemaV1() {
super(FooSchema.class, 1, ImmutableList.of(PersistentFoo.class));
}
@Entity
@Table(name = "foos")
class PersistentFoo implements Serializable {
@Id
@Column(name = "foo_id")
String fooId;
@Column(name = "foo_data")
String fooData;
}
}
object FooSchemaV1 : MappedSchema(
schemaFamily = FooSchema.javaClass,
version = 1,
mappedTypes = listOf(PersistentFoo::class.java)
) {
@Entity
@Table(name = "foos")
class PersistentFoo(
@Id
@Column(name = "foo_id")
var fooId: String,
@Column(name = "foo_data")
var fooData: String
) : Serializable
}
Instances of PersistentFoo
can be manually persisted inside a flow as follows:
PersistentFoo foo = new PersistentFoo(new UniqueIdentifier().getId().toString(), "Bar");
serviceHub.withEntityManager(entityManager -> {
entityManager.persist(foo);
return null;
});
val foo = FooSchemaV1.PersistentFoo(UniqueIdentifier().id.toString(), "Bar")
serviceHub.withEntityManager {
persist(foo)
}
And retrieved via a query, as follows:
getServiceHub().withEntityManager((EntityManager entityManager) -> {
CriteriaQuery<PersistentFoo> query = entityManager.getCriteriaBuilder().createQuery(PersistentFoo.class);
Root<PersistentFoo> type = query.from(PersistentFoo.class);
query.select(type);
return entityManager.createQuery(query).getResultList();
});
serviceHub.withEntityManager {
val query = criteriaBuilder.createQuery(FooSchemaV1.PersistentFoo::class.java)
val type = query.from(FooSchemaV1.PersistentFoo::class.java)
query.select(type)
createQuery(query).resultList
}
Suspendable flow operations cannot be used within the lambda function passed to withEntityManager
. This includes:
FlowSession.send
FlowSession.receive
FlowLogic.receiveAll
FlowLogic.sendAll
FlowLogic.sleep
FlowLogic.subFlow
Restricted control of entity managers
Corda restricts the functions available using the EntityManager
, which is returned by withEntityManager
. This prevents a flow’s underlying database transaction from being tampered with, which would likely lead to errors within the flow.
The withEntityManager
function provides an object that adheres to the EntityManager
interface but with two differences:
getTransaction
returns aRestrictedEntityTransaction
.- All other restricted functions will produce
UnsupportedOperationException
exceptions.
The restricted functions are:
RestrictedEntityManager
:
close()
unwrap()
getDelegate()
getMetamodel()
joinTransaction()
lock() methods
setProperty(propertyName: String?, value: Any?)
RestrictedEntityTransaction
:
rollback()
commit()
begin()
Intermediate database sessions within flows
When you call withEntityManager
, an intermediate database session is created that provides rollback capability without affecting the current transaction.
A withEntityManager
block has the following three outcomes, which are managed by the call and do not require you to manually flush or roll back a flow before calling withEntityManager
:
- Completes successfully: The intermediate session is automatically flushed to the underlying transaction.
- Throws a database error: The intermediate session is automatically rolled back.
- Throws a non-database error: The intermediate session is not flushed to the underlying transaction.
Changes are committed to the database when the transaction is committed.
Handling database errors
A flow can handle database exceptions that occur within a withEntityManager
block without affecting the flow’s underlying database transaction.
withEntityManager
block. Doing so leads to further errors as the flow’s transaction needs to be rolled back.You can handle database errors that occur within a withEntityManager
by catching relevant exceptions. Below are two ways to handle these exceptions:
Around the block:
// try around withEntityManager block
try {
getServiceHub().withEntityManager(entityManager -> {
entityManager.persist(entity);
});
// catch around withEntityManager block
} catch (PersistenceException e) {
// Exception thrown due to constraint violation
getLogger().info("Ok, let's not save this entity 2");
}
// try around withEntityManager block
try {
serviceHub.withEntityManager {
persist(entity)
}
// catch around withEntityManager block
} catch (e: PersistenceException) {
// Exception thrown due to constraint violation
logger.info("Caught the exception!")
}
There is no need for a flush
when catching exceptions around the withEntityManager
block. It automatically triggers a flush
when leaving the block.
withEntityManager
block as it is less likely to lead to unexpected behaviour when interacting with JPA.Inside the block:
getServiceHub().withEntityManager(entityManager -> { entityManager.persist(entity); // try inside withEntityManager block try { // Manually trigger a flush on the intermediate session entityManager.flush(); } catch (PersistenceException e) { // Exception thrown due to constraint violation getLogger().info("Ok, let's not save this entity"); } });
serviceHub.withEntityManager { persist(entity) // try inside withEntityManager block try { // Manually trigger a flush on the intermediate session flush() } catch (e: PersistenceException) { // Exception thrown due to constraint violation logger.info("Ok, let's not save this entity") } }
You must manually trigger a
flush
if the exception is to be caught inside the entity manager. If theflush
is not included, the code above would throw thePersistenceException
instead of catching it.After a database error occurs inside awithEntityManager
block, any executed updates will not be flushed to the underlying database transaction. All changes will be lost as the transaction will be rolled back to the state it had at the beginning of the block.
Manually flushing intermediate database sessions
You need to manually flush
database changes to the underlying database transaction for two reasons:
- Handling database errors: Handle any possible database errors that occur from the
flush
within thewithEntityManager
block. - Survive non-database errors: Keep your database changes even if a non-database error is thrown out of the
withEntityManager
block.
An example of flushing a session to survive a non-database error:
try {
getServiceHub().withEntityManager(entityManager -> {
entityManager.persist(entity);
// Manually trigger a flush on the intermediate session
entityManager.flush();
throw new RuntimeException("Non-database error");
});
} catch (Exception e) {
getLogger().info("I still want to save that entity");
}
try {
serviceHub.withEntityManager {
persist(entity)
// Manually trigger a flush on the intermediate session
flush()
throw RuntimeException("Non-database error")
}
} catch (e: Exception) {
logger.info("I still want to save that entity")
}
flush
the sessions manually in order to survive non-database exceptions, you should keep any code that is likely to cause such errors out of withEntityManager
blocks.Was this page helpful?
Thanks for your feedback!
Chat with us
Chat with us on our #docs channel on slack. You can also join a lot of other slack channels there and have access to 1-on-1 communication with members of the R3 team and the online community.
Propose documentation improvements directly
Help us to improve the docs by contributing directly. It's simple - just fork this repository and raise a PR of your own - R3's Technical Writers will review it and apply the relevant suggestions.
We're sorry this page wasn't helpful. Let us know how we can make it better!
Chat with us
Chat with us on our #docs channel on slack. You can also join a lot of other slack channels there and have access to 1-on-1 communication with members of the R3 team and the online community.
Create an issue
Create a new GitHub issue in this repository - submit technical feedback, draw attention to a potential documentation bug, or share ideas for improvement and general feedback.
Propose documentation improvements directly
Help us to improve the docs by contributing directly. It's simple - just fork this repository and raise a PR of your own - R3's Technical Writers will review it and apply the relevant suggestions.