[ 
https://issues.apache.org/jira/browse/NIFI-14841?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18014519#comment-18014519
 ] 

Praveen Dharam commented on NIFI-14841:
---------------------------------------

We have several custom NiFi processors in our workflow. Here’s the step-by-step 
flow of what happens:

We have a custom Processor P1: This processor receives flow files containing 
comma-separated records, with each file holding multiple records. It processes 
all files in each batch, splits the records into multiple output flow files 
based on certain fields that are concatenated into a string, and assigns that 
string as a flow file attribute.

We run this in a 3-node NiFi cluster, configured with the Partition by 
Attribute load balancing strategy to distribute these files.

Custom Processor P2: This processor consumes the output of P1, where the 
expectation is that each node receives flow files with a unique attribute value.

In NiFi 2.0, using the same input files, we observed that the load was 
distributed fairly evenly across all three nodes (with only minor percentage 
differences). However, in NiFi 2.4, using the same input files, the majority of 
the load is sent to a single node, while the other two nodes receive 
significantly less.

 

> In NiFi 2.4, the load distribution is uneven, whereas it worked correctly in 
> NiFi 2.0
> -------------------------------------------------------------------------------------
>
>                 Key: NIFI-14841
>                 URL: https://issues.apache.org/jira/browse/NIFI-14841
>             Project: Apache NiFi
>          Issue Type: Bug
>            Reporter: Praveen Dharam
>            Priority: Major
>
> In NiFi 2.4, we are experiencing uneven load distribution across nodes when 
> the load balancing strategy is set to Partition by Attribute.
> The primary issue is that the majority of the load is being routed to a 
> single node out of the three in the cluster.
> This imbalance has significantly reduced our throughput during volume 
> testing—an issue that was not present in NiFi 2.0.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to