[ 
https://issues.apache.org/jira/browse/HBASE-29385?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Charles Connell updated HBASE-29385:
------------------------------------
    Description: In HBASE-28770 I introduced new code paths in 
{{AggregateImplementation}} that are activated when the user uses 
{{AsyncAggregationClient}}. The goal of this is to respect HBase's quotas when 
evaluating the user's request. I've been testing this at a large scale. The 
current code checks for quotas exceeded once for every call to 
{{InternalScanner#next()}}. This is not cheap and as a result, calls to 
{{AggregateImplementation}} are tens or hundreds of times slower than they 
otherwise would be. In this ticket I'll do this check less often.  (was: In 
HBASE-28770 I introduced new code paths in {{AggregateImplementation}} that are 
activated when the user uses {{AsyncAggregationClient}}. The goal of this is to 
respect HBase's quotas when evaluating the user's request. I've been testing 
this at a large scale. The current code checks for quotas exceeded once for 
every call to {{ResultScanner#next()}}. This is not cheap and as a result, 
calls to {{AggregateImplementation}} are tens or hundreds of times slower than 
they otherwise would be. In this ticket I'll do this check less often.)

> Very slow performance when using AsyncAggregationClient for large scans
> -----------------------------------------------------------------------
>
>                 Key: HBASE-29385
>                 URL: https://issues.apache.org/jira/browse/HBASE-29385
>             Project: HBase
>          Issue Type: Bug
>            Reporter: Charles Connell
>            Assignee: Charles Connell
>            Priority: Major
>
> In HBASE-28770 I introduced new code paths in {{AggregateImplementation}} 
> that are activated when the user uses {{AsyncAggregationClient}}. The goal of 
> this is to respect HBase's quotas when evaluating the user's request. I've 
> been testing this at a large scale. The current code checks for quotas 
> exceeded once for every call to {{InternalScanner#next()}}. This is not cheap 
> and as a result, calls to {{AggregateImplementation}} are tens or hundreds of 
> times slower than they otherwise would be. In this ticket I'll do this check 
> less often.



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

Reply via email to