fix distribution to not include contrib extensions by default, don't …

[dclim]

…pull the entire AWS SDK bundle

When trying to push the 0.13 RC1 binary to the Apache repository, I was getting rejected, likely due to the file being too large (it was 385MB). While investigating the bloat, I found two major factors:

• As a result of Add maven.exec.xxx.skip option for exec-maven-plugin #6162, all distribution builds were bundling all the contrib extensions
• As a result of Use the official aws-sdk instead of jet3t #5382, we switched to pulling in the entire AWS SDK bundle instead of just the required modules

I fixed the first one by splitting off a separate bundle-contrib-exts profile as previous which can be used to add the contrib extensions; otherwise only the core extensions are bundled.

For the second one, I removed the dependency on aws-java-sdk-bundle and only pulled aws-java-sdk-ec2 and aws-java-sdk-s3. I am not sure if the full bundle was pulled in for any particular reason or just as a convenience.

I also excluded a few larger dependencies from extensions that were already included in the core.

Binary tarball reduced in size from 385MB to 210MB.

FYI, I am still waiting on a response to a JIRA ticket to Apache Legal regarding how the MySQL connector should be packaged: https://issues.apache.org/jira/browse/LEGAL-423. If I don't hear back soon, I'll try to push RC1 (for the third time!) and do an RC2 if they express concerns with our current packaging system. My hunch is that we will be required to separate the offending library (GPL-licensed Java MySQL connector library) from the rest of the MySQL extension.. so if others feel good about it, I wouldn't be opposed to going ahead and making the change.

[b-slim]

@dclim thanks for doing this! I think removing the Dependency on Mysql is the way to go. Hive does that for Mysql, in fact we do not use it that much in the code.

extensions-core/mysql-metadata-storage/src/main/java/org/apache/druid/metadata/storage/mysql/MySQLConnector.java:import com.mysql.jdbc.exceptions.MySQLTransientException;

Same with LZO libs for hadoop/spark.

[gianm]

LGTM. I don't see any discussion in #6162 or #5382 indicating that the changes you mentioned in those two PRs were intentional, and I am in favor of reverting them.

The other changes you made look good too.