ConvertRecord is mostly for changing data formats, not structure. UpdateRecord is more appropriate, but I don't believe this is currently possible, as the "address" field doesn't exist in the input, and we don't currently update the schema in that case. I've filed NIFI-5524 to cover this improvement. You'll also be able to accomplish this with JoltTransformRecord when NIFI-5353 is implemented. In the meantime you can use ConvertRecord to convert from flat CSV into flat JSON (the writer can inherit the record schema), then use JoltTransformJSON to push the fields into the "address" object, here's a spec that will do that: [ { "operation": "shift", "spec": { "*": { "address1": "[#2].address.address_address1", "zipcode": "[#2].address.address_zipcode", "*": "[#2].&" } } } ] ... View more

In addition to Shu's suggestion, you can use UpdateAttribute with Store State set to "Store state locally". This gives you a little more flexibility over the sequence values: you can set an initial value with the "Stateful Variables Initial Value" property, you can increment by any formula you want by specifying it in the UpdateAttribute user-defined property value using NiFi Expression Language, etc. If you must get the value from an external database sequence, you could use a ScriptedLookupService with LookupRecord, you'd write a script to query the database sequence and return the value as the lookup value. I've written a blog post on how to do this with Groovy and PostgreSQL, but it can be done in any supported scripting language against any database (you have to "bring your own SQL" in the script). ... View more

The default implementation is to throw UnsupportedOperationException, but many/most/all of the subclasses override it. Check the reference impl, and since you are removing things, also check removeAndGet(), removeByPattern(), and removeByPatternAndGet() to see if they would help, they might save you the trouble of fetching all keys to remove some of them. ... View more