Rather than creating many FlowFiles with the same Content Claim, refactored content repos' OutputStreams and ClaimWriteCache so that a new ContentClaim is created for each FlowFile. This ensures that we have a content claim offset of 0. The poor performance was due to having to use StreamUtils.skip() in conjunction with the CipherInputStream, which would only skip a max of 511 bytes at a time. By using a separate Content Claim per FlowFile, we no longer need to seek after creating the CipherInputStream
This closes#7363
Signed-off-by: David Handermann <exceptionfactory@apache.org>
(cherry picked from commit 702c6350344e4ae4cd31349747e96c1384017ed4)
* NIFI-11464 Improvements for importing nested versioned flows
- Introduce FlowSnapshotContainer to return root snapshot + children
- Introduce ControllerServiceResolver to extract logic from service facade
- Update resolution logic to correctly consider all services in the hierarchy
- Merge additional parameter contexts and parameter providers from child to parent
- Add unit test for controller service resolver
- Replace use of emptSet/emptyMap with new set/map instance
- Fixed system tests so that they work properly in Clustered version of RegistryClientIT
- Fixed system test - ensure that we wait for processors to become valid before attempting to start them; also added an additional system test around Controller Services in versioned flows
This closes#7095
Signed-off-by: David Handermann <exceptionfactory@apache.org>
Fixed issue in StandardContentClaimWriteCache in which inner OutputStream class did not have an idempotent close() method; as a result, the stream could be written to while already in use for another active FlowFile; fixed bug in ContentClaimInputStream in which skip() method ignored its own BufferedInputStream - this was discovered because it was causing failures in StandardProcessSessionIT; fixed bug in StandardProcessSessionIT in which the length of StandardContentClaim was being doubled because the OutputStream was setting the claim length but that is already handled at a lower level.
This closes#7087
Signed-off-by: David Handermann <exceptionfactory@apache.org>
- Ensure that components are notified that primary node has changed in a background thread instead of the Leader Election thread and activate/deactivate the thread in the case of Processors so that they can be viewed in the UI and terminated
- Fixed system tests that would fail intermittently because they did not wait for node disconnection to complete and did not properly switch the client to look at the connected node before checking cluster status
This closes#7052
Signed-off-by: David Handermann <exceptionfactory@apache.org>
NIFI-11192: If a failure is encountered when changing the version of a flow from 1 version to another, attempt to rollback the changes instead of just failing with the flow in a bad state
Signed-off-by: Matthew Burgess <mattyb149@apache.org>
This closes#6981
Correcting EmbeddedQuestDbStatusHistoryRepositoryForComponentsTest
Correcting TestQueryNiFiReportingTask
Signed-off-by: Matthew Burgess <mattyb149@apache.org>
This closes#6869
NifiRegistryFlowRegistryClient defines the PropertyDescriptor PROPERTY_URL name as small case "url". The map bases on the name property of the PropertyDescriptor object. Here searching with uppercase value of "URL" causes the map lookup to fail and cause a NPE later on. Therefore, it is changed as "url"
NIFI-10787 - Added constant for property descriptor "url" in NiFiRegistryFlowMapper to make it more clear.
NIFI-10787 - Added change to unit test.
NIFI-10787 - Updated unit test to validate that NiFi registry url is being set and retrieved.
Signed-off-by: Nathan Gough <thenatog@gmail.com>
This closes#6655.
