ignite-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Gianfranco Murador (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (IGNITE-708) Need to remove background partition exchange
Date Sun, 13 Sep 2015 14:28:45 GMT

    [ https://issues.apache.org/jira/browse/IGNITE-708?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14742498#comment-14742498

Gianfranco Murador commented on IGNITE-708:

>From what I could see,  the refresh partition in case of  timeout is redundant, since
each node has a refresh partition which runs at regular intervals in the method body() of
GridCachePartitionExchangeManager retrieved from the background process ExchangeWorker. Thus
eliminating the refreshPartition event in class ResendTimeoutObject should not have any impact
on the operation, but an increase in performance in when we avoid sending redundant message.
I will proceed with the creation of a pull request. 

> Need to remove background partition exchange
> --------------------------------------------
>                 Key: IGNITE-708
>                 URL: https://issues.apache.org/jira/browse/IGNITE-708
>             Project: Ignite
>          Issue Type: Task
>            Reporter: Yakov Zhdanov
>            Assignee: Gianfranco Murador
>            Priority: Blocker
>              Labels: datagrid
> Now every node sends its partition map to cache coordinator (which is the oldest node
in topology) and coordinator spreads full partition map to every node in topology. This happens
for each cache separately. This seems to take place even if there were no changes to local
partition maps. Given we guarantee communication message delivery this background process
seems to be an overkill.
> Exchange should happen only if any changes took place.
> After dynamic cache start has been introduced, we can have significant amount of live
caches at some point of app lifecycle and app may suffer from  background exchange which is
obviously not a requirement (and may be never has been the one).

This message was sent by Atlassian JIRA

View raw message