[
https://issues.apache.org/jira/browse/NIFI-11450?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17950700#comment-17950700
]
David Handermann commented on NIFI-11450:
-----------------------------------------
Thanks for raising these questions [~Nissim Shiman].
I agree the general expectation should be that {{autoResumeState}} is set the
same on all nodes, given that cluster voting determines the accepted status of
the flow configuration.
There might be a place to consider logging a warning about ignoring the
property setting in cases where a cluster vote overrides the value, but on the
other hand, warnings are easily ignored.
Another more involved option could be to have cluster nodes compare their
autoResumeState as part of the voting process, but that would not be trivial.
The end result would probably be similar, where some nodes behave one way, and
some behave another way.
My initial thought is to add some documentation noting that this must be set to
the same value across all cluster nodes. As it is a manual change, this seems
like it could be a reasonable solution, avoiding the inherent complexities of
some alternatives.
> Node respecting nifi.flowcontroller.autoResumeState when joining cluster
> ------------------------------------------------------------------------
>
> Key: NIFI-11450
> URL: https://issues.apache.org/jira/browse/NIFI-11450
> Project: Apache NiFi
> Issue Type: Bug
> Components: Core Framework
> Affects Versions: 1.19.1
> Environment: Nifi version 1.19.1. Linux. JDK11+
> Reporter: Phil Lord
> Assignee: Nissim Shiman
> Priority: Critical
>
> Mistakenly tried to connect a node to an active cluster that had the
> following setting: nifi.flowcontroller.autoResumeState=true .
> The flow.xml/json were both removed completely prior to attempting to connect
> node to the cluster, then started the node. The node connected to the
> cluster without issue, however all components/controllerServices on that node
> were in a stopped state(due to the property mentioned above).
> Is this correct behavior given that the flow that was inherited from the
> cluster had components started/etc. I would think the node should either
> fail to join the cluster/start?
> This was running Nifi 1.19.1... unclear if it's resolved in latest release.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
