[
https://issues.apache.org/jira/browse/NIP-37?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18100554#comment-18100554
]
David Handermann commented on NIP-37:
-------------------------------------
Thanks for proposing this improvement [~pvillard], it should provide a better
interface surface that clarifies that state of the Node.
The only adjustment I recommend is an additional enumerated type for
unclustered or standalone, as opposed to implying to from the empty status.
> Expose Node Connection State through NodeTypeProvider
> -----------------------------------------------------
>
> Key: NIP-37
> URL: https://issues.apache.org/jira/browse/NIP-37
> Project: NiFi Improvement Proposal
> Issue Type: Improvement
> Reporter: Pierre Villard
> Priority: Major
>
> h2. Motivation
> *NodeTypeProvider* currently exposes cluster status through several boolean
> methods:
> * *isConfiguredForClustering()* indicates whether the NiFi instance is
> configured as a cluster node.
> * *isClustered()* indicates whether the node is connected or attempting to
> connect.
> * *isConnected()* indicates whether the node is fully connected.
> These booleans do not provide enough information to describe the node
> lifecycle accurately. In particular, components can be loaded and validated
> while a node is joining a cluster.
> During clustered startup, the Cluster Coordinator marks the local node as
> *CONNECTING* before the cluster flow is loaded. However,
> *FlowController.isClustered()* is not set to true until after flow
> synchronization completes. Setting it earlier is not appropriate because
> *FlowController.setClustered(true, ...)* also starts cluster-related behavior
> such as heartbeating, leader-election participation, load balancing, and
> Remote Process Group reinitialization.
> As a result, a component loaded during flow synchronization can observe:
> * isConfiguredForClustering() == true
> * isClustered() == false
> * isConnected() == false
> From these values, the component cannot distinguish a standalone or
> disconnected node from a node that is actively connecting to a cluster.
> Components that need to validate behavior based on cluster topology can
> therefore produce incorrect transient validation results during startup.
> NiFi already models this lifecycle using {*}NodeConnectionState{*}, with
> states such as {*}CONNECTING{*}, {*}CONNECTED{*}, {*}DISCONNECTING{*}, and
> {*}DISCONNECTED{*}. This proposal makes that state available through
> {*}NodeTypeProvide{*}` instead of requiring components to infer it from
> multiple booleans.
> h2. Scope
> Add the following method to {*}NodeTypeProvider{*}:
> {code:java}
> default Optional<NodeConnectionState> getNodeConnectionState() {
> return Optional.empty();
> }{code}
> The default implementation preserves compatibility with existing
> implementations of {*}NodeTypeProvider{*}.
> The proposed semantics are:
> * *Optional.empty()* means that the provider does not expose a cluster
> connection state, including providers representing standalone NiFi instances.
> * A present value represents the current connection lifecycle state of the
> local node.
> * *CONNECTING* means that the node is joining the cluster but has not
> completed connection.
> * *CONNECTED* means that the node has completed the connection process.
> * Other existing *NodeConnectionState* values retain their current meanings.
> *NodeConnectionState* currently belongs to the
> nifi-framework-cluster-protocol module, which is in nifi. To expose it from
> {*}NodeTypeProvider{*}, move *NodeConnectionState* into the nifi-api artifact
> while retaining its existing package name, enum constants, ordering, and
> serialization behavior.
> Retaining the fully qualified class name avoids requiring source changes in
> existing NiFi framework code that uses the enum.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
