Moved start up logic of the various node components out of their c’tors and into “start” methods, which are called from Node.start(). The components themselves are created in the Node’s c’tor with minimal initialisation logic.
Certain things are not immediately available at construction time, which are instead given to the components at start time in an orderly fashion:
* Certs from the node’s key store and trust store
* The network parameters, and thus access to things like maxMessageSize and the contracts whitelist
* A running database - several components were doing database stuff their c’tors
* The node’s NodeInfo, and thus access to things like identities and addresses
The messaging service couldn’t be created in the Node’s c’tor due to initialisation issues with MockNode. This should be fixed in a later commit.
The loading of all node infos in the start method was unnecessary, both for the changePublisher and _loadDBSuccess, and the setting of _registrationFuture was incorrect.
They're prone to cause flaky tests due to the "allocated" ports already being taken by the system when eventually needed. Replaced usages with PortAllocation.Incremental.
Affected unit tests made into integration tests to avoid any issues in the future when unit tests are made to run in parallel.
* CORDA-1743: Remove `isDebug = true` from Node driver to speed-up integration tests execution.
* CORDA-1743: Undo removal of "isDebug = true" from the test that specifically checks for this condition.
* CORDA-1743: Address input from @shamsasari
* ENT-2014 Deletes of NodeInfo can fail to propagate leading to infinite retries
ENT-1880 Move identity key generation to network registration process
(cherry picked from commit c3ac203)
* [CORDA-1634] Destroy child processes when parent exits.
* Add comment.
* Register Shutdownhook for processes regardless of whether the Driver was initialized with
* Add comment.
* Revert "Add comment."
This reverts commit a5e78c379f.
* Add comment.
* Add shutdown hook in ShutdownManager.registerProcessShutdown.
* Initialize the ShutdownManager with a shutdown hook to ensure that is called.
* Add comment.
As reported in [CORDA-1609](https://r3-cev.atlassian.net/browse/CORDA-1609),
`CordaRPCClientConfiguration.default` is not accessible from Java since
`default` is a reserved keyword.
As part of the refactor made in #2831, `CordaRPCClientConfiguration` went
from being a data class to an interface with a backing implementation of
type `CordaRPCClientConfigurationImpl`.
This resulted in Java users having to rewrite code that was on the form:
```java
final CordaRPCClient client = new CordaRPCClient(
nodeAddress, CordaRPCClientConfiguration.DEFAULT
);
```
to something like this:
```java
final CordaRPCClient client = new CordaRPCClient(
nodeAddress, CordaRPCClientConfiguration.Companion.default()
);
```
However, this does not work. The user would get a compilation error because
`default` is a reserved keyword in Java.
Since `CordaRPCClientConfiguration` has been made an interface, there is no
easy way of introducing a static final field on the interface from Kotlin.
Consequently, I've changed this back to using a `class` with a static field
named `DEFAULT` instead of the static method `default()`.
It should be noted that `default()` / `DEFAULT` is currently only used
internally to pass in default values in `CordaRPCClient.kt` and
`CordaRPCClientUtils.kt`. That said, it is exposed as part of our API
surface and consequently shouldn't be broken.
The latter means that in the above example, the user would actually not
have to provide the parameter at all:
```java
final CordaRPCClient client = new CordaRPCClient(nodeAddress);
```
As can be seen from the definition of `CordaRPCClient`:
```kotlin
class CordaRPCClient private constructor(...) {
@JvmOverloads
constructor(
hostAndPort: NetworkHostAndPort,
configuration: CordaRPCClientConfiguration = CordaRPCClientConfiguration.DEFAULT
) : this(hostAndPort, configuration, null)
```
The mentioned [refactor](7a077e76f0 (diff-0948c125db93a22263eb81eaf3161c17R65))
did not make it into the 3.1 release, so from an API-stability perspective,
this change can be applied without affecting our commitment to a
backwards compatible API..
* filter by contract state in _trackBy
* write tests to check that _trackBy is filtering the states correct and tidy up filtering functions
* remove un needed function
* add change log message for filtering unrelated ContractStates from trackBy
* CORDA-1494: Remove isTimed check from flow hospital when handling timeout exceptions - the timeout might expire just after the TimeFlow has finished and the staff member would throw an exception.
Increase the default flow timeout value in mock network so retries don't happen – this has cause test flakiness.
It will re-run automatically from last checkpoint on node restart, allowing the opportunity to resolve the issue, something required when dealing with contract constraint failures.
* Remove all notion of message level retry.
* Introduce randomness into de-duplication IDs based on the session rather than the flow, in support of idempotent flows.
* CORDA-1494: Re-enable notarisation retries in the new state machine manager.
The original message-based retry approach does not work well with the new
flow state machine due to the way sessions are handled. We decided to move
the retry logic to flow-level: introduce RetryableFlow that won't have
checkpoints persisted and will be restarted after a configurable timeout
if it does not complete in time.
The RetryableFlow functionality will be internal for now, as it's mainly
tailored for the notary client flow, and there are many subtle ways it can
fail when used with arbitrary flows.
JPA/Hibernate entities need to impose the correct NULL/NOT NULL constraints on the database - whatever these correct values actually are.
API change: net.corda.core.schemas.PersistentStateRef fields (index and txId) are now non-nullable. Rationale: The fields were always effectively non-nullable - values were set from non-nullable fields of other objects. The class is used in context of database table Primary Key of for other entities and a database already imposes those columns as non-nullable (even if JPA annotation nullable=false was absent).
* CORDA-1510 - Allow Doorman and NetworkMap to be configured independently
Currently only one compatabilityZoneURL can be specified, however the
two services can be run on as separate servers. Allow nodes to be
configured in this manner
* Partial review comments
* Review comments
* review comments
node-api now depends on this module and upcoming changes will use this as well rather than having to depend on node-api.
EnumEvolveTests.deserializeWithRename and EnumEvolveTests.multiOperations are temporarily ignored since their test resources can't be regenerated due to bugs.
It's no longer used as we've switched over to AMQP for RPC calls so
remove it from everywhere and only use it for checkpointing
* Wire up demo bench post Kryo removal
* Test Fixes
* rebase and fix tests
* Test Fix
* wip
* revert changes to api now we don't need to add annotations
* Client and server support for amqp
* Observable (and supporting) serialisers
Unit Tests
* Fixing tests
* Test fixes
* CORDA-847 - Update api doc with additon of @CordaSerializable annotation
* TestFixes
* review comments
* TestFixes
* Test Fix
* Test Fix
* Test Fix
* Test Fix
* Test Fix
* Test Fix
* TestFix
* Test Fix
* Review Comments
* Sync public API's (and associated internal helpers) between Enterprise and OS.
* Add previous default constructor explicitly.
* Keep the API checker happy!
* Revert addition of ENT makeTestDataSourceProperties() function.
