Andrew Wong has posted comments on this change. ( )

Change subject: WIP KUDU-16 pt 3: add per-scan-token limits

Patch Set 1:

> Patch Set 1:
> (1 comment)

After some discussion with Thomas from the Impala team, I think the consensus 
is that this API isn't that useful. Right now, Impala can't really take 
advantage of this top-level LimitPerToken API because the scan limit isn't 
known at the time of token building (although there are likely ways around this 
through updating Impala's source).

The ideal API would be one where a limit could be set per grouop of tokens and 
the tokens would be able to coordinate with each other to satisfy the exact 
limit. This is difficult to do, so it's probably off the table for now.

That said, the limit is definitely known during the re-hydrating call 
DeserializeIntoScanner(), so Impala should be able to use the API defined in 
 in a similar way to that in this patch, i.e. specifying the overall limit per 
scanner such that l * t rows are returned across all scanners and returning the 
first l, since this is still a perf win.

To view, visit
To unsubscribe, visit

Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I655f07f10a9a99e9766402729361de97e929136a
Gerrit-Change-Number: 9923
Gerrit-PatchSet: 1
Gerrit-Owner: Andrew Wong <>
Gerrit-Reviewer: Andrew Wong <>
Gerrit-Reviewer: Dan Burkert <>
Gerrit-Reviewer: Kudu Jenkins
Gerrit-Reviewer: Todd Lipcon <>
Gerrit-Comment-Date: Mon, 09 Apr 2018 21:43:58 +0000
Gerrit-HasComments: No

Reply via email to