You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@nifi.apache.org by "Mark Payne (Jira)" <ji...@apache.org> on 2021/03/23 13:51:00 UTC
[jira] [Commented] (NIFI-8353) When node is offloaded, it may still
receive data from load-balanced connections
[ https://issues.apache.org/jira/browse/NIFI-8353?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17307078#comment-17307078 ]
Mark Payne commented on NIFI-8353:
----------------------------------
Not only should an offloaded node not receive data via load-balanced connections, but any node that is disconnected should not receive data via load-balanced connections either. Currently, it appears that we are not checking if a node is connected to the cluster before sending data to the node. This can also result in a node taking a very long time to complete offloading.
> When node is offloaded, it may still receive data from load-balanced connections
> --------------------------------------------------------------------------------
>
> Key: NIFI-8353
> URL: https://issues.apache.org/jira/browse/NIFI-8353
> Project: Apache NiFi
> Issue Type: Bug
> Components: Core Framework
> Reporter: Mark Payne
> Assignee: Mark Payne
> Priority: Blocker
> Fix For: 1.14.0
>
>
> Consider a simple flow: GenerateFlowFile (9 FlowFiles per batch, each 10 KB, scheduled once per second) -> UpdateAttribute.
> The connection between the two processors is configured with Round Robin load balancing.
> Start GenerateFlowFile. Wait a few seconds for some data to queue up.
> Disconnect Node 2. Immediately offload Node 2. Refresh cluster page to see that the node is offloaded.
> Navigate to Node 2's UI. Sometimes it will show data queued up. This is because after the node was offloaded, the connection from Node 1 was still transferring data to Node 2. Node 1 won't start any additional transactions to send data to Node 2, but it will complete the already-started transactions. So the node can be offloaded and end up with data queued.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)