[ 
https://issues.apache.org/jira/browse/IGNITE-27760?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Roman Puchkovskiy updated IGNITE-27760:
---------------------------------------
    Description: 
The client subscribes to LT (Logical Topology) changes. The following could 
happen:
 # Node version 1 joins the LT
 # The node gets restarted in k8s, so it starts as version 2 on another IP
 # Node version 2 joins the LT

As a result, the LT will contain 3 events: v1 joins (with old address), v1 
leaves, v2 joins (with new address).

When a node (maybe the same node) starts, its topology-aware client will start 
getting the events.
 # It sees v1 join and attempts a subscription, which caused a network 
invocation, which fails as old address does not have the node anymore. These 
attempts will be retried indefinitely
 # It sees v1 leave, but it does not stop the attempts from step 1, so they 
will never stop

We should terminate such subscription attempts.

> Topology aware Raft client disregards node leaving LT
> -----------------------------------------------------
>
>                 Key: IGNITE-27760
>                 URL: https://issues.apache.org/jira/browse/IGNITE-27760
>             Project: Ignite
>          Issue Type: Bug
>            Reporter: Roman Puchkovskiy
>            Assignee: Roman Puchkovskiy
>            Priority: Major
>              Labels: ignite-3
>
> The client subscribes to LT (Logical Topology) changes. The following could 
> happen:
>  # Node version 1 joins the LT
>  # The node gets restarted in k8s, so it starts as version 2 on another IP
>  # Node version 2 joins the LT
> As a result, the LT will contain 3 events: v1 joins (with old address), v1 
> leaves, v2 joins (with new address).
> When a node (maybe the same node) starts, its topology-aware client will 
> start getting the events.
>  # It sees v1 join and attempts a subscription, which caused a network 
> invocation, which fails as old address does not have the node anymore. These 
> attempts will be retried indefinitely
>  # It sees v1 leave, but it does not stop the attempts from step 1, so they 
> will never stop
> We should terminate such subscription attempts.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to