[
https://issues.apache.org/jira/browse/IGNITE-13179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17448059#comment-17448059
]
Aleksey Plekhanov edited comment on IGNITE-13179 at 11/23/21, 6:03 PM:
-----------------------------------------------------------------------
[~korlov], first problem (distribution trait propagation) was already resolved
by IGNITE-14973.
What about:
{quote}Also Limit must be propogated (copied) under Exchange to reduce
unnecessary fetches.
{quote}
I'm not sure it will be very efficient since we already request the next
batches of rows only until the limit is reached, so the only overhead here is
the buffer size. But we can also make one more optimization: requests from
{{LimitNode}} source no more than {{limit}} rows (currently we request the same
amount of rows as requested by downstream). WDYT?
I've raised a pull request to show what I mean.
was (Author: alex_pl):
[~korlov], first problem (distribution trait propagation) was already resolved
by IGNITE-14973.
What about:
{quote}Also Limit must be propogated (copied) under Exchange to reduce
unnecessary fetches.
{quote}
I'm not sure it will be very efficient since we already request the next
batches of rows only until the limit is reached, so the only overhead here is
the buffer size. But we can also make one more optimization: requests from
{{LimitNode}} source no more than {{limit}} rows (currently we request the same
amount of rows as requested by downstream). WDYT?
> Calcite integration. Fix traits at the IgniteLimit
> --------------------------------------------------
>
> Key: IGNITE-13179
> URL: https://issues.apache.org/jira/browse/IGNITE-13179
> Project: Ignite
> Issue Type: Improvement
> Reporter: Taras Ledkov
> Assignee: Konstantin Orlov
> Priority: Major
> Labels: calcite2-required, calcite3-required
> Time Spent: 10m
> Remaining Estimate: 0h
>
> Current implementation of {{IgniteLimit}} doesn't support distribution trait
> propagation.
> Limit must require singleton distribution.
> Now we cannot reproduce invalid behavior (see IGNITE-12819) and IgniteLimit
> isn't placed under the Exchange node.
> Also Limit must be propogated (copied) under Exchange to reduce unnecessary
> fetches.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)