[jira] [Commented] (GEODE-6576) DB connection rebalancing improvement

["Mario Ivanac (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/GEODE-6576?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16806502#comment-16806502 ] 

Mario Ivanac commented on GEODE-6576:
-------------------------------------

Hi,

 

I can provide you with more details:
 * 3 servers / 10 clients
 * geode 1.8
 * 8 cores / 14 GB per server
 * Linux (RHEL 7.5)
 * 6 GB heap/ 8 GB off-heap
 * Single hop enabled
 * Shenandoah GC

 

but we have identified source of problem, and it is C++ Geode Native driver.

 

We have observed increased latency spikes that are happening in 5 minute intervals.

From what we see the parameter load-conditioning-interval has default of 5 minutes and this is aligned with the behavior.

After checking the code, we can see that Geode Native has a manage connections thread which is running and checking all the connections.

This seems to be the source of the issue.

> DB connection rebalancing improvement
> -------------------------------------
>
>                 Key: GEODE-6576
>                 URL: https://issues.apache.org/jira/browse/GEODE-6576
>             Project: Geode
>          Issue Type: Improvement
>          Components: native client
>            Reporter: Mario Ivanac
>            Assignee: Mario Ivanac
>            Priority: Major
>
> When using min=max=120 connections, and about 2000 TPS it is observed, that every 5 minutes  (load-conditioning-interval) we get significant delays . But I suspect you could reproduce it at any configuration, if you look for increased delays when driver rebalances connections (load-conditioning-interval). 



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)