You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ignite.apache.org by "Vyacheslav Koptilin (Jira)" <ji...@apache.org> on 2022/10/18 12:30:00 UTC

[jira] [Updated] (IGNITE-16609) Thin 3.0: Design Cluster Discovery for Java thin client

     [ https://issues.apache.org/jira/browse/IGNITE-16609?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Vyacheslav Koptilin updated IGNITE-16609:
-----------------------------------------
    Fix Version/s: 3.0.0-beta2
                       (was: 3.0.0-beta1)

> Thin 3.0: Design Cluster Discovery for Java thin client
> -------------------------------------------------------
>
>                 Key: IGNITE-16609
>                 URL: https://issues.apache.org/jira/browse/IGNITE-16609
>             Project: Ignite
>          Issue Type: New Feature
>          Components: thin client
>    Affects Versions: 3.0.0-alpha4
>            Reporter: Igor Sapego
>            Priority: Major
>              Labels: ignite-3
>             Fix For: 3.0.0-beta2
>
>
> We should design and implement a mechanism for thin clients that will provide them with ability to discover nodes in cluster.
> My thoughts on the matter: 
> If we implement this approach (which should be a default I belive) then initial addresses provided to thin client by a user should only be used to establish the first connection to a node of the cluster. After that, thin client should request full list of active nodes of the cluster and use them instead of the initial list to establish new connections.
> The problems that should be considered during desing and implementation steps:
> - How should clients be notified about changes in cluster (should clients send requests for that or should server issue notifications);
> - How should clients handle a situation when they can not reach cluster nodes by addresses provided by server.
> - How should clients behave on connection to nodes from different clusters.
> - How should clients handle different states of the cluster lifecycle.
> These problems should be probably addressed in separate tickets. In this case, those ticket should be created and linked to the current ticket.



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