ignite-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Alexey Goncharuk (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (IGNITE-9682) Update full partition map in parallel
Date Tue, 23 Oct 2018 12:52:00 GMT

    [ https://issues.apache.org/jira/browse/IGNITE-9682?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16660605#comment-16660605

Alexey Goncharuk commented on IGNITE-9682:

[~oleg-ostanin], the change looks good, however, I am concerned that we may consume all threads
which may lead to a deadlock. Recently we've introduced a new method, {{U.doInParallel(int
parallelismLevel, ...)}} which does a parallel code execution in limited number of threads.
You can see an example usage in {{GridCacheProcessor.prepareStartCaches(...)}}. 
Can you change your code similarly to have at least two system threads available?

> Update full partition map in parallel
> -------------------------------------
>                 Key: IGNITE-9682
>                 URL: https://issues.apache.org/jira/browse/IGNITE-9682
>             Project: Ignite
>          Issue Type: Improvement
>            Reporter: Oleg Ostanin
>            Assignee: Oleg Ostanin
>            Priority: Major
> I've noticed that we are updating full partition map for each cache sequentially and
during such update we are unable to run much of other tasks. Hence there should be some free
threads which we can use for parallel updating. Sometimes it takes 20-30 seconds to update
a map for 100+ caches before rebalancing can even start:
> {noformat}
> [2018-09-24 16:38:49,016][INFO ][sys-#219|#219] Received full message, will finish exchange
[node=4d6b5921-0a61-4bfa-bf11-b8a1b8332dce, resVer=AffinityTopologyVersion [topVer=19, minorTopVer=2]]
> ...
> [2018-09-24 16:39:01,217][INFO ][sys-#219|#219] Finish exchange future [startVer=AffinityTopologyVersion
[topVer=19, minorTopVer=2], resVer=AffinityTopologyVersion [topVer=19, minorTopVer=2], err=null]
> {noformat}

This message was sent by Atlassian JIRA

View raw message