[ 
https://issues.apache.org/jira/browse/IMPALA-7780?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16667720#comment-16667720
 ] 

Pooja Nilangekar commented on IMPALA-7780:
------------------------------------------

I have faced the issue of different estimates earlier. After asking around and 
poking though the dataload, I found that the file sizes and hence the estimates 
may vary between different instances of data load. You might end up using 
less/more disk space to load the exact same data on the same/ another machine 
with identical specs. So I am not sure rebase would actually solve this issue. 

> Rebase PlannerTest expected output for estimates, errors
> --------------------------------------------------------
>
>                 Key: IMPALA-7780
>                 URL: https://issues.apache.org/jira/browse/IMPALA-7780
>             Project: IMPALA
>          Issue Type: Improvement
>          Components: Frontend
>    Affects Versions: Impala 3.0
>            Reporter: Paul Rogers
>            Priority: Trivial
>
> The front-end includes the {{PlannerTest}} test which works by running a 
> query, writing the plan to a file, comparing selected parts of the file to 
> expected results, and flagging if the results differ.
> A plan includes some things we test (operators) and some we do not (text of 
> error messages, value of memory estimates). Over time the expected and actual 
> files have drifted apart. Example:
> {noformat}
> Expected:    partitions=1/1 files=2 size=54.20MB
> Actual:      partitions=1/1 files=2 size=54.21MB
> {noformat}
> While the tests still pass (because we ignore the parts which have drifted), 
> it is a pain to track down issues because we must learn to manually ignore 
> "unimportant" differences.
> This ticket asks to "rebase" planner tests on the latest results, copying 
> into the expected results file the current "noise" values from the actual 
> results.



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

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to