When a `UnexpectedFlowEndException` or a `FlowException` is received the
peer that the exception was thrown from will be added to the stacktrace.
This is due to it being easier to see and a field that developers
are much less likely to override.
A nullable field `peer` has been added to `FlowException` and
`UnexpectedFlowEndException`. This is read later on (when peer info
is not available) to append the peer info to the stacktrace.
`ExceptionSerialisingRpcOpsProxy` was removing information about the
original exception in dev mode. This code has been removed. Although
there is no check in `ExceptionSerialisingRpcOpsProxy` for dev mode
(and also due to it being about serialising),
`ExceptionMaskingRpcOpsProxy` will handle the removal of exception data
in non dev mode (production mode).
* Follow up changes to error reporting around failed flows
* Have FinalityDoctor report stack trace
* Revert changes to the DumpHistoryOnErrorInterceptor
* Initial version of new(old) dialog that won't print a stacktrace for rpc exceptions.
* Decoupled CordaVersionProvider. Moved common files to common-logging to lower dependencies on the node explorer.
* Removed unused import and duplicate documentation comment.
* Moved error code rewrite policy in the new common/logging module according to PR review.
* Removed extra line.
* Updated log4j configurations with new package name where logging policies will be contained.
* Included common-logging module with cliutils.
* setting non existant base directory now yields suitable error message
* corrections to missing base directory message and comments added for clarity
* added check for valid base directory before node loggin and bootstrapping
* removed uneeded import
* quantum computer joke back by popular demand and not back at the same time
* CORDA-2813 Add DbException handler to terminate process when the DB goes away or gives up.
Also add a helper to terminate the node and use it instead of calling Runtime.halt() in random places.
* CORDA-2813 Address threading/blocking concerns raised during review
* Tests for custom registry restrictions
* ENT-3121 restrict custom serialisation
* Remove redundant code
* Only count declared annotations
* Check annotation on superclasses, remove annotation from ByteArray
* Forbid custom serialization of primitive types
* Remove @CordaSerializable from another class that is always handled by custom serialisation
* Add log warnings to aid diagnosis of custom serialization issues
* Remove another annotation
* Remove another annotation
* Remove another annotation
* Remove another annotation
* Fixup api-current
* Fixup api-current
* KDocs on exceptions
Addressed review comments - removed nested if and added parameter name for the usage of requiresCertificates.
Removed unnecessary import and white space.
Normally, these requests are left unacknowledged in the MQ broker to allow recovery via installing the missing CorDapp. If this is not applicable then the request be dropped (and the other side informed of the error).
While the node is starting up, we now check for the presence of the certificates directory. This allows us to print out an easily understandable error message if the directory is not present. An exception is made for devMode, as devMode will result in the directory being created in any case.
* ENT-3057: Log hibernate warns and errors in different log
If a hibernate error occurs (deadlock, for example) that would cause a flow to be sent to the hospital, hibernate logs the warnings and errors before we do. This results in duplication in the logs, and pollutes the log. To solve this, we create a new log appender named diagnostic-{node-name}.log and log any org.hibernate messages of warn and above to that file. This way, messages are not lost, which means that the information can be retrieved if need be.
* Corrected indentation of comment (changed tab to space)
* Updated node-administration document to mention diagnostic logging change
* Fixed integration test. It was breaking because it was fetching the first log file in the folder, assuming there would be only one. This assumption is now invalid because the diagnostic log file that was introduced. Two tests were found that used similar logic to find a log file to examine, hence both were corrected to look for log files beginning with "node"
* Updated documentation as per review comments.
Add a cache for notary identities in the PersistentIdentityService. This
solves a reported problem where notary identity lookup fails if its network
map entry is missing, which results in an exception when trying to insert
a state into the vault after recording a transaction.
Test if we have access to the logging path (baseDirectory/logs) before attempting to write to them. This allows us to shut down gracefully with an easily understandable error message.
Without doing this, the log4j2 will attempt to access the logPath when it first uses the logger in the call() function.
* CORDA-2629: Provide a better error msg when notary type misconfig
If a notary service is misconfigured with the type (either validating or not validating) stored in the node's configuration not matching that advertised in the network map cache, the notary will throw an exception and fail on startup.
Previously, this misconfiguration would result in an exception being thrown when attempting to notarise a transaction. This change results in the exception being thrown at node startup and the node operator being aware of the misconfiguration earlier.
* Corrected exception message when the notary does not have a party
* Changed exception message to include configured and advertised values.
Fixed unit test that was failing because of the new check.
* Plumb through the crlCheckSoftFail configuration option to bridge manager
* Add crlCheckSoftFail test to bridge manager and fix equivalent proton wrapper test
* Update documentation and set the node configuration default to true
* Revert default change and clarify consequences of setting option to false
* Remove NodeConfiguration default to leave only AMQPConfiguration default
There's a bug with the ServiceLoader which leaks a file handle to the app jar on shutdown. This causes an issue if a mock node is restarted in Windows. To avoid the problem completely we no longer overwrite any existing jars, as the jar to be copied will be same anyway.
(cherry picked from commit 0038a86481)
Only log a line about looking for notary implementations if at least notary implementation is found.
It might be unnecessary and confusing to log when no implementations are found.
Replaced TrustedAuthorityNotaryService with SinglePartyNotaryService in comment
* CORDA-2665: Updated OwnableState relevancy check is now gated on target version 4
https://github.com/corda/corda/pull/3789 changed the relevancy check of OwnableState to include the participants list in addition to the owner. This however breaks existing apps which assume (in their vault query) an OwnableState is recorded to the vault if-and-only-if the owner matches.
* CORDA-2665 Don't switch to new behaviour on target version upgrade.
When an RPC client disconnects from the RPC server, the log entry created now has its log level set to info rather than warning. It is perfectly normal for an RPC client to disconnect - only the RPC client knows if it was intentional.
* CORDA-2669 - pendingFlowsCount not in public API
Reintroduce `pendingFlowsCount` to public API (as deprecated). Advise
to use the `gracefulShutdown` command in the shell instead.
* CORDA-2669 - Add pendingFlowsCount to api-current.txt
* ENT-3053 Database connection pools leaking memory on every checkpoint. Flip in the thread local from the thread into the fiber.
* Back port to OS (needs some gradle changes) and added TODO, ability for it to avoid erroring if not using Hikari.
* Review feedback to remove warning.
* CORDA-2634 - Fix default currentTargetVersion
* CORDA-2634 - Include abstract flow classes in CorDapp class scanning
* CORDA-2634 - Run test for target platform version 4
To allow rolling upgrades where members of a BN may not all upgrade at the same time, the FinalityHandler is now always enabled, and not disabled if no old apps are installed.
Previously any attempt to use the FinalityHandler in such a scenerio would send the flow to the hospital, where manual intervention would be required to recover the transaction.
* Use the attachments classloader to deserialize contract states in migrations
* Added some comments to explain serialisation behaviour and how tests work.
* Add debug log to indicate when attachment classloading has failed.
* Use a servicesForResolution to load states for compatibility with notary changes and contract upgrades
* Add test case to cover notary change transactions
* Address review comments
* Change logging message in MigrationServicesForResolution
* Read the network-parameters file if there is nothing in the database
* Update documentation and provide a warning if there are many states.
* Handle case where file containing network parameters is missing
* Improve logging if attachments are not available and the file path doesn't exist
* Get logs the right way around
* Fix build error in tests
* Improve error when transaction deserialisation fails and move migrations for finance to contracts CorDapp
* Revert move of migrations and errors thrown from CordaRPCOps
* Ensure VaultQueryException is thrown from vault queries and remove unused import
* Improve error reporting from VaultQueryException
* Fix API break
* Fix vault query test failure due to exception change
CORDA-2595 - Fix test and api.
CORDA-2595 add test
CORDA-2595 fix tests
CORDA-2595 fix test and address code review comments
CORDA-2595 address code review comments
* First pass at fixing 2563.
* In memory KMS now maps keys to IDs.
* CreateDatabaseAndMockServices now creates a persistent key management service and a can take a persistent identity service, so now the external id mapping works for mock services.
* * Created a helper for mock services which allows the creation of a mock services with persistent identity management service key management service and vault.
* MockNode now uses persistent key management service - not sure why it didn't do before?
* * MockNode now uses BasicHSMKeyManagementService
* Updated api-current file
* Little fix required after rebase to master.
* Fixed broken test.
* Added informative error messages to UnsupportedOperationExceptions thrown by E2ETestKeyManagementService.
* Removed redundant private constructor for mock services from api-current.txt.
* Addressed Rick's comments.
* Create a new custom migration for populating the state_party table and the relevancy_status column within the vault_states table when migrating from V3 to V4.
* Add a corresponding unit test suite.
* Small refactorings to expose data required by the migration: the isRelevant function in the NodeVaultService, and the node's legal name from the configuration.
Previous implementation was in LedgerTransaction and focused only on contract classes,
but every package matters.
Also fixes some exception types and does misc refactorings.
Take out a useless parameter from a method that was added to the public
API, document it. Add some comments explaining more about why we are
looking up attachment versions in WireTransaction.toLedgerTransaction.
* CORDA-2547: WIP Migrate old notary schema correctly.
Modify liquibase scripts to rename the old committed state table instead of creating a new one.
* Rename liquibase change sets
* Change 3.3 baseline before Liquibase.
* Aligin with ENT, orginal node-notary.changelog-v1.xml moved to node-notary.changelog-v2.xml and it's compatible with ENT (which already has changes from v2).
* Aligin with ENT, orginal node-notary.changelog-v1.xml moved to node-notary.changelog-v2.xml and it's compatible with ENT (which already has changes from v2) - adding v2 to master file.
* Ensure setting up Liquibase on pre-existing pre-Liquibase database
* Add FetchParametersFlow
* No downgrade parameters in ResolveTransactionsFlow
Make sure that parameters in the transaction
graph are ordered (this is to prevent the downgrade attack, when the
malicious notary and participants sign transaction that shouldn't be
notarised otherwise). We ensure that by checking that epochs of network
parameters in the transaction chain are ordered.
* Addressed some minor items from RP review feedback.
* Refactoring following rebase from master.
* Address RP PR review comments (round 2)
* Addressed a couple of minor PR review points.
* Renaming of unit tests and cleanup.
* Changes discusses with RP to ensure Network Param checking is applied at txn verify time + resolve order checking gated on existence of tagged NPs in txn and associated minimum platform version.
* Do not fail on missing ServiceHub impl + return nothing if txn not NP tagged.
* Unify HistoricNetworkParametersStorage and
NetworkParametersStorageInternal
* SignedDataWithCert implements NamedByHash
* Cleanup
* Move parameters ordering check to signed transaction resolution
* Fixes after merge, address comments
* Address Andrius comments
This includes all of the cash selection logic, JSON support using jackson and a bunch of utilities which are not relevant to contract verification. The exception to this are Interpolator which has been moved to the IRS demo, and PhysicalLocationStructures.kt which is now in explorer.
* Allow for duplicate (contract class, version) signed attachments in devMode.
* Code clean-up.
* Fix compilation error in test code.
* Additional gating/warning and added Unit test for development mode behaviour.
* Remove locks around database access in AppendOnlyPersistentMap and introduce
a unit test that checks that known deadlock scenarios of the old version
are avoided.
* Fix Deadlock unit test
* Add some extra latching to try and make timing less fragile. Can never be perfect though.
* Review feedback, and some thread safety fixes.
* Start pooling classpath scanning
Quickly patch synchronisation of attachment class loader cache. Needs a revisit but more complicated due to DJVM.
Annotate away for DJVM
Take ClassGraph utils into their own file so we can exclude for DJVM
Clean up a little
* Daemonize the threads
* Seems to be some concurrency problems with use of ClassGraph. Using a mutex for now to work around.
Merged three tests into one to avoid repeating the same node startups. The assertions the original tests made can be done in one test without any contamination.
* Removed `restart node successfully with suspended flow` as it duplicates `TraderDemoTest#Test restart node during flow works properly`
* Removed the need for a notary
This is only testing the internal cordappForClasses method which is used for internal testing. This method is covered by other unit and integration tests.
CORDA-2497 fix
* Added test to show the fix working.
* Now backwards compatible.
* Refactored out some duped code.
* Added better explanations for what's going on.
* Fixed test which was failing due to the serializationEnvRule problem.
* Addressed Tudor's review comments.
The column 'uploader' was marked as not updatable entity so any update wasn't propagated to the database when the attachment was uploaded again and the uploader field was changed to a trusted one. Also 'saveOrUpdate' removed, the entity is already managed, saveOrUpdate is for detached/new entity.
Upgrade from node 3.0 to 4.0 fails to create versions of whitelisted JARs from networkParameters - read parameters from file at first as in Corda 3.0 there no relevant table, then try from the table.
As this is migration, the code will run only once on each node, she increased log level to info for messages. Tested using https://r3-cev.atlassian.net/browse/R3T-1549
* Update changelog with changes to vault queries
* Move choice of relevancy state to QueryCriteria
* Ensure relevancy default is correct for all query types
* Remove superfluous imports
* Update documentation for queries using state relevancy
* Fix merge damage
* Revert change to use RELEVANT as the default and update docs
* Revert missed instance of relevancy change
* Rewrite of upgrade notes for vault state query updates
* Address review comments
There were two issues:
* The original "sealing violation: can't seal package net.corda.nodeapi" issue was due to the isolated CorDapp containing some code in the net.corda.nodeapi namespace. This has been moved to the isolated namespace.
* The test was not correctly creating the second transaction with the dummy command
* Add whitelists and custom serializers from cordapps to serialization context
* Remove changes in TransactionBuilder, add caching
* Add whitelists and custom serializers from cordapps to serialization context
* Remove changes in TransactionBuilder, add caching
* Address comments
* Increase node memory for SIMM integration test
* Cache only serialization context
* Increase integ test timeout
* Fix API breakage
* Increase max heap size for web server integ test
* Move classloading utils from separate module to core.internal
* Adjust heap size for more integ tests
* Increase time window for IRS demo transactions
* Fix determinator
* Add parameter in core-deterministic
* Stub out class-loading method for DJVM
* Pass states to record through to transaction resolution
* Add a test case
* Add comment indicating why states are always added in tx resolution
* Update observer node documentation
The version of contract attachments that are whitelisted should be read from NetworkParameters.whitelistedContractImplementations.
It use the lattes network map from db with the highest epoch.
* Fixed bug in state pointer search and added tests.
* Blacklisted problematic package.
* Addressed Shams' comments.
* Addressed round two of comments.
* Fixed another bug whereby the DFS gets stuck in an infinite loop.
The API has been reverted to be completely ABI compatible with V3, and the small changes that were made to the wire format in https://github.com/corda/corda/pull/4260 have also been reverted.
Corda Node ensures a given contract class and version can be sourced from only one signed and trusted Attachment (JAR).
An attempt to import a signed JAR as a trusted uploader (or promote to be trusted) with a class and version already present in the other trusted Attachment will raise DuplicateContractClassException.
Minor fixes to Hibernate Attachment Query parser (original query to select attachment without signers would always return no attachments)
Move Raft and BFT-Smart notaries back into node to preserve backwards compatibility.
* Allow overriding full node config when using internal mock network parameters.
* Make BFT-Smart notary start up in prod mode as well
* Move raft & bftsmart notaries to net.corda.notary.experimental package
* Make sure Raft notary handles reference state edge cases correctly.
* Make sure BFT-Smart notary handles reference state edge cases correctly.
* Include notary schemas in node internal schemas
* Undo Raft notary table schema changes to maintain compatibility.
* Add cordapp code signing dev key to production blacklist.
* Remove code used to generate the actual PK hash (it is now attached to the JIRA story)
* Remove wiki page reference.
* Remove the non redundant cash transfer.
* ENT-2923 - remove db access code from the verification thread pool
* Remove worker pool for tx verification and disable db access.
* Address code review comments
* Reduce the operation of ip autodetect to nodes specifying localhost, rather than any node with a 192.168.x.x or 10.x.x.x address. Also, catch the common user error of specifying 0.0.0.0 in p2pAddress.
* Fix test that uses invalid config
Apps may depend on the previous behaviour of catching Throwable rather than just Exception. Better to not risk this break and instead provide a helper to throw Errors.
Also using Consumer to avoid ugly usage in Java for doOnSuccess and doOnFailure.
* Update the documentation for trackBy to better indicate that updates are not filtered according to the query criteria
* Add a test that shows the behaviour, which is skipped for now until a solution is implemented.
