[
https://issues.apache.org/jira/browse/NIFI-4638?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16285295#comment-16285295
]
Netanel Bitan commented on NIFI-4638:
-------------------------------------
Hi!
I am also experiencing this bug, and after doing some research I have some
information to add:
1. This happens in many types of processors.
2. The disorder is always between events that are generated by the processor
itself, such as Content Modified or Attributes Modified, and events that are
generated by the framework, such as Rout and Drop.
3. The disorder happens in 1% of the lineages (out of ~250000 lineages)
4. In 99% of the disorders the difference between the timestamps of the two
events is 1 millisecond, but sometimes it's more, up to half a second .
> Drop events disordering
> -----------------------
>
> Key: NIFI-4638
> URL: https://issues.apache.org/jira/browse/NIFI-4638
> Project: Apache NiFi
> Issue Type: Bug
> Reporter: Omer Hadari
>
> Drop events that are generated by terminated relationships alongside other
> event types are sometimes registered before them. This happens since the drop
> event’s timestamp is lower by a millisecond compared to the timestamp of the
> other event, even though the ordinal is higher. This causes the lineage graph
> to look as if the other event comes out of the drop event.
> The one millisecond difference and the fact that this bug is sporadic
> suggests a rollover to the next millisecond when registering the event before
> the drop event.
> For example this was encountered in the split json processor, when the
> original flowfile receives a fragment count and then dropped. This could be a
> bug in the processor itself, though it seems to be more generic.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
