[
https://issues.apache.org/jira/browse/NIFI-13613?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17870223#comment-17870223
]
Night Gryphon commented on NIFI-13613:
--------------------------------------
[~exceptionfactory] this thing is critical as flow with ghosted components make
NiFi cluster fail. Not the moment such flow imported but if any node need to
reconnect - the whole cluster fails.
Nodes can't reconnect back to cluster with ghosted components and at the same
time i can't remove that ghosted components until i restore cluster. So to
overcome this i has to shutdown the WHOLE cluster than node by node clean
ghosted components from each node flow separately than reassemble the cluster.
Such behavior cause significant downtime.
Here is the steps to reproduce the problem (cluster of two NiFi nodes + NiFi
Registry):
* import flow with old components from registry causing to get ghosted
components. For example org.apache.nifi.processors.standard.JoltTransformJSON
* restart one of NiFi nodes - node will fail to connect back to cluster and
NiFi will shutdown on that node
* try to remove ghosted components from flow to let failed node to reconnect -
NiFi will block delete operation as it can't sync with failed node
now we have failed cluster which can't be reassembled back
* remove failed node from cluster configuration
* delete process group with ghosted components
* try to restart failed node again - it will again fail to reconnect cluster
and shutdown
> Update class names when import older flow from registry
> -------------------------------------------------------
>
> Key: NIFI-13613
> URL: https://issues.apache.org/jira/browse/NIFI-13613
> Project: Apache NiFi
> Issue Type: Bug
> Components: Core Framework
> Affects Versions: 2.0.0-M3
> Environment: Ubuntu 22.04
> Open JDK 21
> NiFi Registry 2.0.0.M3
> Reporter: Night Gryphon
> Priority: Critical
>
> I'm trying to migrate from NiFi 1.23 to 2.0.0 using NiFi Registry.
> When i import flow to 2.0.0 i get some missing components due to package
> names changes for some components like JOLT.
> The issue/bug is having such flow prevent cluster nodes to restart/rejoin.
> Once node is shutdown or disconnected from cluster it can't come back as it
> can't sync flow with master node due to missing components and so node just
> shutdown. Furthermore at the master node i can not delete or fix the corrupt
> process group as some nodes are unavailable and can't rejoin cluster.
> So to recover i has to delete nodes from cluster, delete process group and
> than rebuild cluster.
> To fix this issue class names has to be checked at the flow import and
> replaced according to updates in 2.0.0.
> Also flow sync at join cluster has to somehow deal with missing components
> which are exist in master flow and actually can be displayed and manually
> fixed.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
