[jira] [Commented] (HIVE-18668) Really shade guava in ql

["Sergey Shelukhin (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HIVE-18668?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16393532#comment-16393532 ] 

Sergey Shelukhin commented on HIVE-18668:
-----------------------------------------

Hmm... I think this actually broke the shading.
The string is group ID and artifact name, which in ql pom is the way it was.
I can name it foo:bar and it will still work if it matches the dependency IDs in the pom.

With this patch:
{noformat}
$ jar tf ./ql/target/hive-exec-3.0.0-SNAPSHOT.jar | grep -i stopwatch | grep -v apache
Picked up JAVA_TOOL_OPTIONS: -Djava.awt.headless=true
jodd/datetime/JStopWatch.class
{noformat}

Without this patch:
{noformat}
$ jar tf ./ql/target/hive-exec-3.0.0-SNAPSHOT.jar | grep -i stopwatch | grep -v apache
Picked up JAVA_TOOL_OPTIONS: -Djava.awt.headless=true
jodd/datetime/JStopWatch.class
com/google/common/base/Stopwatch$1.class
com/google/common/base/Stopwatch.class
com/google/common/util/concurrent/RateLimiter$SleepingStopwatch$1.class
com/google/common/util/concurrent/RateLimiter$SleepingStopwatch.class
{noformat}


Looks like presence of guava actually breaks Druid tests.
I am going to revert this for now because it makes Hive pick random guava jars (version across Hadoop, Tez etc is a big mess), so it cannot be a long term fix for Druid anyway. Druid tests need to be fixed in some other manner.

cc [~bslim]

> Really shade guava in ql
> ------------------------
>
>                 Key: HIVE-18668
>                 URL: https://issues.apache.org/jira/browse/HIVE-18668
>             Project: Hive
>          Issue Type: Bug
>          Components: Druid integration
>            Reporter: Zoltan Haindrich
>            Assignee: Zoltan Haindrich
>            Priority: Major
>             Fix For: 3.0.0
>
>         Attachments: HIVE-18668.01.patch
>
>
> After HIVE-15393 a test started to fail in druid; after some investigation it turned out that ql doesn't shade it's guava artifact at all...because it shades 'com.google.guava' instead 'com.google.common'



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