[
https://issues.apache.org/jira/browse/NIFI-1645?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Joseph Witt updated NIFI-1645:
------------------------------
Description:
When using the delimited lines feature to send data to Kafka such that a large
set of lines that appear to be one 'flowfile' in NiFi is sent as a series of
1..N messages in Kafka the mechanism of asynchronous acknowledgement can break
down whereby we will receive acknowledgements but be unable to act on them
appropriately because by then the session/data would have already been
considered successfully transferred. This could in certain/specific conditions
mean failed acknowledgements would not result in a retransfer.
The logic this processor supports for creating child objects to address
failed/partial segments is extremely complicated and should likely be rewritten
to be greatly simplified. Instead the SplitText feature should be used to
create more manageable chunks of data over which if any segment is ack'd as a
failure then the whole thing is failed and thus can be retransmitted. Always
best to enable the user to prefer data loss or data duplication on their own
terms.
> When using delimited data feature PutKafka ack'd ranges feature can break
> -------------------------------------------------------------------------
>
> Key: NIFI-1645
> URL: https://issues.apache.org/jira/browse/NIFI-1645
> Project: Apache NiFi
> Issue Type: Bug
> Reporter: Oleg Zhurakousky
> Assignee: Oleg Zhurakousky
> Fix For: 0.7.0
>
>
> When using the delimited lines feature to send data to Kafka such that a
> large set of lines that appear to be one 'flowfile' in NiFi is sent as a
> series of 1..N messages in Kafka the mechanism of asynchronous
> acknowledgement can break down whereby we will receive acknowledgements but
> be unable to act on them appropriately because by then the session/data would
> have already been considered successfully transferred. This could in
> certain/specific conditions mean failed acknowledgements would not result in
> a retransfer.
> The logic this processor supports for creating child objects to address
> failed/partial segments is extremely complicated and should likely be
> rewritten to be greatly simplified. Instead the SplitText feature should be
> used to create more manageable chunks of data over which if any segment is
> ack'd as a failure then the whole thing is failed and thus can be
> retransmitted. Always best to enable the user to prefer data loss or data
> duplication on their own terms.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
