Github user mattyb149 commented on the issue:
https://github.com/apache/nifi/pull/2430
Ugh that's true, I'm not a fan of dynamic keys. Since the schema would be
generated by the reporting task, then we could create a schema for the example
above, but then each flow file would have its own schema, and even worse, each
flow file would have only one metric, so at that point it's not really
conducive to record processing. As an alternative we could convert the output
(before or after JSON conversion) to an altered spec with a consistent schema
definition.
For the Ambari reporting task, since Ambari is expecting this format, then
fine; if we keep the same spec in this new reporting task for consistency, then
I'd hope to see a template on the Wiki using the new reporting task with a
JoltTransformJSON processor to do the aforementioned transformation, along with
including an AvroSchemaRegistry that contains the schema definition for the
files coming out of the JoltTransformJSON processor. This would allow us to
keep a consistent (standard) defined format (albeit non-schema-friendly), but
offer a well-known solution to prepare the data for record-aware processors.
Thoughts?
---
