question Re: [Nifi 2.4] - Round robin queues stucked just after migration in Support Questions

[Nifi 2.4] - Round robin queues stucked just after migration

OlivierCS — Fri, 12 Dec 2025 11:00:31 GMT

Hello,

We have a 3 nodes cluster. We migrate our system from Nifi 1.23.2 to Nifi 2.4 with a stop at Nifi 1.28.1 as recommended in migration docs.

Our flows were working as expected for the part 1.23.2 -> 1.28.1. We prepared the migration to Nifi 2 as explain in docs (remove variables and deprecated processors, modified CRON to remove the 7th digit...). Tests were OK.

We then upgrade from 1.28.2 to Nifi 2.4. Just after this upgrade, our flows containing queues with round robin strategy getting stucked.

One example, we ingest a XML file in the primary node (let's call it node0), do some stuffs, no problem. We then split the file by lines and send each lines to an EvaluateXPath to analyse the data. This queue between the SplitXML and EvaluateXPath has a round robin strategy to improve perfs. Just after the upgrade from 1.28.1 to Nifi 2.4 only one third of the lines were processed the lines that should be sent to the node1 et node2 are not processed and stay in the queue. Only reccords assigned to the primary node are processed (node0).

Our admin did some tests to check if it was a network issue or certifs issue. It seems not as he succeed to execute some telnet and SLL command from node0 to node1 or node2. Zookeeper connection between nodes is OK as we manage to modify flows and having the modifications reported everywhere. After analysing logs he just found that there is an acknowlegement issue when node0 is trying to send flowfiles to node1/node2.

We didn't find any known bug in Nifi 2.4 regarding the round robin strategy. Is there anyone elso who faced this kind of issue and managed to solve it.

Thank you very much for your help,

Olivier

Re: [Nifi 2.4] - Round robin queues stucked just after migration

VidyaSargur — Fri, 12 Dec 2025 15:27:05 GMT

@OlivierCS, Welcome to our community! To help you get the best possible answer, I have tagged in our NiFi experts @MattWho, @haridjh who may be able to assist you further.

Please feel free to provide any additional information or details about your query, and we hope that you will find a satisfactory solution to your question.

Re: [Nifi 2.4] - Round robin queues stucked just after migration

vafs — Fri, 12 Dec 2025 21:03:41 GMT

Hello @OlivierCS,

Thanks for being part of this community.

I think this applies to https://issues.apache.org/jira/browse/NIFI-14841.
This looks like there are some issues known on NiFi 2.4 about load distribution.
According to the Jira, it's solved in 2.6.
Do you have this NiFi standalone or it's CFM? If standalone, you may take a look and consider using that newer version.

Maybe other members on the community have more ideas to share.

Re: [Nifi 2.4] - Round robin queues stucked just after migration

OlivierCS — Mon, 15 Dec 2025 08:37:48 GMT

Hello Andrés,

Yes we had a look at it, in fact it seems it's really solved with 2.7 with https://issues.apache.org/jira/browse/NIFI-14959.

I add a look at the code fix provided and well it seems to concern really this partition by attribute. By the way, when we saw this defect, I tried another test.

I create a dummy attributs and assign randomly a value 0 - 10 and then tried the partition load by this atrribute. This was even worst than the load balance strategy(I had less reccords processed than the third I had with load balace strategy). If it was the same kind of issue, I was expecting to have roughly the same amount of processed reccords. I say roughly because in this second test I had perhaps 10% of the reccords processed vs 33% with the load balance strategy.

In addition, we are not big fans of pushing to our Prod right now a version (2.7.1) which has been released 3 days ago only. That was why we were searching if someone else faced the issue or if someone else was using this version with load balance strategy without having issue. Just to try to narrow down where the issue comes from: nifi 2.4 or our platform.

Anyway thank you very much for your answer Andrés,

Best regards,

Olivier