[ 
https://issues.apache.org/jira/browse/DRILL-4609?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16003332#comment-16003332
 ] 

Karthikeyan Manivannan commented on DRILL-4609:
-----------------------------------------------

It is unlikely that this is a JVM bug. JDK 7's runtime/JIT-compilation policies 
might just be hiding a Drill bug. JVM bugs( like mis-compiles, incorrect class 
loading) usually result in a crash.
All that said, given that this is a 1-bit error, it is possible that this is a 
JVM bug.  

You can eliminate the role of the JIT compiler by trying to repro the bug with 
the -Xint( interpreter only, will slow down execution significantly) JVM option.
If you hit the bug with -Xint then you can be fairly certain that this is not a 
JVM bug.





> Select true,true,true from ... does not always output true,true,true
> --------------------------------------------------------------------
>
>                 Key: DRILL-4609
>                 URL: https://issues.apache.org/jira/browse/DRILL-4609
>             Project: Apache Drill
>          Issue Type: Bug
>          Components: Client - CLI, Query Planning & Optimization, Storage - 
> Writer
>    Affects Versions: 1.5.0, 1.6.0
>         Environment: Linux Redhat
> tested in cluster (hdfs) and embedded mode
>            Reporter: F Méthot
>
> Doing a simple "select true, true, true from table" won't output 
> true,true,true on all generated rows.
> Step to reproduce.
> generate a simple CSV files:
> {code:sql}
>       for i in {1..1000000}; do echo "Allo"; done > /users/fmethot/test.csv
> {code}
> Open a new fresh drill CLI.
> Just to help for validation, switch output to CSV: 
> {code:sql}
>       alter session set `store.format`='csv' 
> {code}
> generate a table like this:
> {code:sql}
>        create table TEST_OUT as (select true,true,true,true from 
> dfs.`/users/fmethot/test.csv')
> {code}
> Check content of /users/fmethot/test.csv
> You will find false values in there!
> If you generate another table, on the same session, the same way, chances are 
> the value will be fine (all true). We can only reproduce this on the first 
> CTAS run. 
> We came to test this select pattern after we realize our custom boolean UDF 
> (as well as the one provided in Drill like "ilike") were not outputting 
> consistent deterministic results (same input were implausibly generating 
> random boolean output). We hope that fixing this ticket will also fix our 
> issue with boolean UDFs.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Reply via email to