Mark Payne created NIFI-2039:
--------------------------------
Summary: Provide ability to get InputStream from FlowFile without
using Process Session callback
Key: NIFI-2039
URL: https://issues.apache.org/jira/browse/NIFI-2039
Project: Apache NiFi
Issue Type: Improvement
Components: Core Framework
Reporter: Mark Payne
Fix For: 1.0.0
Provide a mechanism to obtain an InputStream for a FlowFile directly from the
ProcessSession without having to provide an InputStreamCallback.
Currently, we often have processors that create an AtomicReference so that the
value can be set within a callback and then the value must be obtained using a
get() once the callback returns. This is very un-intuitive. Additionally, we
don't have the ability to return an InputStream to other code that may required
to run in a callback. For instance, when using Calcite, Calcite wants a
callback, essentially, that provides an Enumerator of data. We can't currently
support this well because we can't create an InputStream and return it.
We need to add the following method to ProcessSession:
InputStream read(FlowFile flowFile) throws ProcessException;
The InputStream that is returned must be tracked by ProcessSession such that
when the session is committed or rolled back, the stream is closed if it hasn't
been already. All of the logic that currently takes place in
ProcessSession.read() after the callback returns must be done on the closing of
the new InputStream that is returned.
The MockProcessSession should throw an Exception if commit() is called without
the InputStream being closed, ideally providing the stack trace of where the
stream was obtained and where commit() was called. This is to ensure that the
developer is being a good consumer, but the StandardProcessSession doesn't
really need to do this, as we can simply close the InputStream so that it can
no longer be consumed.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
