[jira] [Commented] (PHOENIX-1428) Keep scanner open on server and pace by client instead of spooling

["James Taylor (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/PHOENIX-1428?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15088435#comment-15088435 ] 

James Taylor commented on PHOENIX-1428:
---------------------------------------

Thanks for the patch, [~samarthjain]. I believe we also do wrapping with a SpoolResultIterator in AggregatePlan as well which we need to handle (maybe I missed that in your patch?):
{code}
    private ParallelIteratorFactory wrapParallelIteratorFactory () {
        ParallelIteratorFactory innerFactory;
        QueryServices services = context.getConnection().getQueryServices();
        if (groupBy.isEmpty() || groupBy.isOrderPreserving()) {
            innerFactory = new SpoolingResultIterator.SpoolingResultIteratorFactory(services);
        } else {
            innerFactory = new OrderingResultIteratorFactory(services);
        }
        if (parallelIteratorFactory == null) {
            return innerFactory;
        }
        // wrap any existing parallelIteratorFactory
        return new WrappingResultIteratorFactory(innerFactory, parallelIteratorFactory);
    }
{code}

Also, should the ScanUtil.isPacingScannersPossible() method call isRenewingLeasesEnabled too, as the feature may be disabled.

> Keep scanner open on server and pace by client instead of spooling
> ------------------------------------------------------------------
>
>                 Key: PHOENIX-1428
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-1428
>             Project: Phoenix
>          Issue Type: Bug
>            Reporter: James Taylor
>            Assignee: Samarth Jain
>              Labels: SFDC
>         Attachments: PHOENIX-1428.patch
>
>
> Instead of spooling a batch of results for all chunked scans to the client, keep the scan open and pace it through pre-fetching. This will perform much better for a full table scan with a LIMIT = 1, as this case currently will run a scan for every guidepost, each returning a single row.
> [~lhofhansl]



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)