[
https://issues.apache.org/jira/browse/NIFI-2592?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16930752#comment-16930752
]
Mark Payne commented on NIFI-2592:
----------------------------------
One gotcha that we'll want to avoid here is the following pattern, which we
have in `AbstractListProcessor`, `ListHDFS`, and possibly others:
# In `onPropertyModified` check if the modified property is one that would
necessitate a clearing of the state (for example, the directory being
monitored). If so, set a `resetState` flag.
# In an `@OnScheduled` method, check the `resetState` flag and if it's set,
clear the cluster state. Then clear the flag.
This works currently because when the processor is started, the `@OnScheduled`
runs on all nodes at the same state. If we changed this to instead only execute
when the node becomes Primary Node, what will happen is that the property will
be modified, and the flag will be set. Then the processor will be started and
run for some time, perhaps days or even months. Then, when the Primary Node
changes (due to the node being restarted or whatever), another node will be
elected the Primary Node. Then, the `@OnScheduled` method will be invoked, and
it will check its flag, which is already set. It will end up clearing state
that was amassed over the time that the other node was running. This is
essentially the issue that was occurring in NIFI-6677.
To avoid causing this to happen, I would recommend introducing a new concept on
the `PropertyDescriptor` such that the developer can indicate
`clearStateOnChange(Scope...)` that indicates that if the value of the property
is changed, then state (LOCAL, CLUSTER, or BOTH, according to the provided
argument) must be cleared. Then, the framework should handle clearing the state
as appropriate.
> Processor still runs @OnScheduled method when it's running on a non-primary
> node
> --------------------------------------------------------------------------------
>
> Key: NIFI-2592
> URL: https://issues.apache.org/jira/browse/NIFI-2592
> Project: Apache NiFi
> Issue Type: Bug
> Components: Core Framework
> Affects Versions: 1.0.0
> Reporter: Jeff Storck
> Priority: Major
>
> If a processor is scheduled to run on the primary node only, the instances of
> the processor running on non-primary nodes still have their methods annotated
> with @OnScheduled invoked. The framework should not invoke the @OnScheduled
> methods if the processor isn't supposed to run on that node.
> To reproduce this, set up a cluster with the nodes running on the same
> server. The ListenHTTP can be scheduled to be run on the primary node only,
> but other instances of the processor on non-primary nodes will still try to
> bind the port, resulting in exceptions.
--
This message was sent by Atlassian Jira
(v8.3.2#803003)
