Known issues in Flow Management
Learn about the known issues in Flow Management clusters, the impact or changes to the functionality, and the workaround.
Learn about the known issues and limitations in Flow Management in this release:
- NiFi UI becomes unavailable when a bulletin is generated at Flow Controller level
- When a bulletin is generated at the Flow Controller level, such as for reporting tasks
or a disconnected node, the NiFi UI becomes unavailable and gets stuck with a loading
spinner. The UI remains inaccessible until the bulletin at the Flow Controller level is
cleared (a bulletin persists for 5 minutes after being generated).
The issue is a result of NIFI-11433 and is fixed in NIFI-11560. The fix is not included in this release.
- Incomplete Ranger policy for NiFi metrics in Cloudera Manager
- To have Cloudera Manager properly reflect the NiFi metrics for the NiFi service, the Flow NiFi access policies in Ranger needs to be updated to include the "nifi" group.
- KafkaRecordSink puts multiple records in one message
- All the records are sent as a single Kafka message containing an
array of records.
For more information, see NIFI-8326.
- NiFi Atlas reporting task does not work after data lake upgrade from light to medium
- After you upgrade your data lake from light to medium scale, the data lake machine hostname and IP address will change. As the Atlas reporting task uses Atlas and Kafka server hostnames, after the upgrade the wrong hostnames will prevent NiFi to report into Atlas.
- InferAvroSchema may fail when inferring schema for JSON data
- In Apache NiFi 1.17, the dependency on Apache Avro has been upgraded to 1.11.0. However, the InferAvroSchema processor depends on the hadoop-libraries NAR from which the Avro version comes from, causing a NoSuchMethodError exception. Having well defined schemas ensures consistent behavior, allows for proper schema versioning and prevents downstream systems to generate errors because of unexpected schema changes. Besides, schema inference may not always be 100% accurate and can be an expensive operation in terms of performances.