[
https://issues.apache.org/jira/browse/GEODE-4148?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Bruce Schuchardt updated GEODE-4148:
------------------------------------
Description:
Clients periodically send ClientReplacementRequests to the locator to rebalance
the number of client connections across the cluster. When multiple clients are
getting rebalanced, the lack of any buffering can cause clients to move between
servers without changing the loading. For example, if there are an odd number
of clients on an even number of servers, the locator may ping-pong clients
between the server that has one more client than the others.
The locator should wait until some threshold to move clients, to avoid
ping-ponging and unnecessary connection churn.
was:
Clients periodically send a {{ClientReplacementRequest}}s to the locator to
rebalance the number of client connections across the cluster. When multiple
clients are getting rebalanced, the lack of any buffering can cause clients to
move between servers without changing the loading. For example, if there are an
odd number of clients on an even number of servers, the locator may ping-pong
clients between the server that has one more client than the others.
The locator should wait until some threshold to move clients, to avoid
ping-ponging and unnecessary connection churn.
> Locator threshold before rebalancing clients
> --------------------------------------------
>
> Key: GEODE-4148
> URL: https://issues.apache.org/jira/browse/GEODE-4148
> Project: Geode
> Issue Type: Improvement
> Components: client/server, locator
> Reporter: Galen O'Sullivan
>
> Clients periodically send ClientReplacementRequests to the locator to
> rebalance the number of client connections across the cluster. When multiple
> clients are getting rebalanced, the lack of any buffering can cause clients
> to move between servers without changing the loading. For example, if there
> are an odd number of clients on an even number of servers, the locator may
> ping-pong clients between the server that has one more client than the others.
> The locator should wait until some threshold to move clients, to avoid
> ping-ponging and unnecessary connection churn.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
