Some processors may not do anything to do the flow file, meaning they
don't modify the attributes and don't modify the content. An example
would be a routing processors like RouteOnAttribute. I believe routing a
flow file does not produce events because otherwise there are tricky
cases like when a processor has a self-loop and may be retrying an error
tons of times, you would get all these unnecessary route events back to
self.
Source and destination processors are responsible for
generating the send/receive/fetch events themselves because NiFi itself
can't detect when a processor talks to an external system. So a
processor like FetchHBaseRow should always generate a fetch event,
assuming it calls session.getProvenanceReporter().fetch(flowFile,
transitUri), which it does.
Other processors that are in the
middle of the flow and are modifying flow files don't have to worry
about generating events because the framework can detect when a flow is
written to, or when it's attributes are modified, so those events will
be generated automatically.
There will typically be some delay as
to when the events are searchable and viewable... the provenance repo is
made up of a write-ahead-log and lucene indexes, I believe the lucene
indexes are updated at certain checkpoints like every 1-2 mins, and
those indexes are what back the displays in the UI.
