Do you have information on:
a) compatibility issues.
b) performance?

The most significant performance improvement was added in 1.3.0, allowing a Put to contain multiple values (https://github.com/OpenTSDB/asynchbase/commit/bb9168cda1551db4712c1c584cd140a66ec17422) - in 1.2.0, the same operation would require many separate Puts. It also lets us use coprocessors on the region server, which for us at least is a functional requirement.

WRT bugs, we've experienced frequent ZooKeeper connection issues when using Flume 1.3.0 and Hadoop 1.0.3. Recompiling Flume with AsyncHBase 1.3.2 fixed this issue, although I've yet to identify exactly what changeset fixed this (I'm assuming it's one of the Netty changes).

The ZK connection issues might have been due to ZOOKEEPER-1437. I am concerned about the dependencies it introduces. Looks like it will pull in new version of Netty, Guava etc. Let me take a look at the dependencies it is pulling in by doing a build.

It looks like ZK is being downgraded from 3.4.4 to 3.3.4. I am not sure why this is and whether this affects HBase. I will need to look at hbase and see if this change in ZK version is a concern.

Outputs of mvn dependency:tree. Do you know why this is happening:

With asynchbase-1.3.2:

[INFO] +- org.apache.flume.flume-ng-sinks:flume-ng-hbase-sink:jar:1.4.0-SNAPSHOT:compile
[INFO] |  \- org.hbase:asynchbase:jar:1.3.2:compile
[INFO] |     +- com.stumbleupon:async:jar:1.2.0:compile
[INFO] |     +- org.apache.zookeeper:zookeeper:jar:3.3.4:compile
[INFO] |     \- org.slf4j:jcl-over-slf4j:jar:1.6.4:runtime

With aynchbase-1.2.0 (current trunk):

[INFO] +- org.apache.flume.flume-ng-sinks:flume-ng-hbase-sink:jar:1.4.0-SNAPSHOT:compile
[INFO] |  \- org.hbase:asynchbase:jar:1.2.0:compile
[INFO] |     +- org.jboss.netty:netty:jar:3.2.7.Final:compile
[INFO] |     +- com.stumbleupon:async:jar:1.2.0:compile
[INFO] |     +- org.apache.zookeeper:zookeeper:jar:3.4.4:compile
[INFO] |     \- org.slf4j:jcl-over-slf4j:jar:1.7.2:runtime

ZOOKEEPER-1437 is indeed the same issue we had. Regretfully I hadn't spotted the change from 3.4.4 to 3.3.4 (curse those similar numbers!)

Looking deeper, AsyncHBase has always had a dependency on ZooKeeper 3.3.1 (in 1.2.0) or 3.3.4 (1.3.0 and higher) - see https://github.com/OpenTSDB/asynchbase/blob/master/third_party/zookeeper/include.mk - which is making me wonder why 3.4.4 is being pulled in at all.

Ah, the AsyncHBase POM explicitly defines a minimum version. Still doesn't explain why it pulled in 3.4.4 before and 3.3.4 now.

David Johnston Do you have any updates on this? It might be a good idea to open an issue with asynchbase project to look into this. You should post the mvn dependency:tree output which I have posted above.

We can bump it up to 1.4.0. I think at this point, we can just ignore the ZK dependencies.

OK, looks like what was happening was that asynchbase went from using dependency ranges to specific dependency versions. So I think at this point we can upgrade to the latest version.

[~cdcttr], do you want to still look into this one? If yes, please update to Asynchbase 1.4.1

Going to AsyncHBase 1.4.0 may resolve FLUME-2028 as well.

Yes, it would - since Asynchbase-1.2.0 did have open versioning for zk, which this version gets rid of. I am assigning this to myself since David has not responded for a couple days.

Trivial patch, skipping rb. Ran full build with both profiles - works fine.

+1 lgtm

Pushed to trunk & flume-1.4 branches.

Thanks for the patch Hari!

Integrated in flume-trunk #407 (See https://builds.apache.org/job/flume-trunk/407/)
FLUME-1688. Bump AsyncHBase version to 1.4.1. (Revision c2f559482f60cea4899dc0dc57a88c7087755823)

Result = FAILURE
mpercy : http://git-wip-us.apache.org/repos/asf/flume/repo?p=flume.git&a=commit&h=c2f559482f60cea4899dc0dc57a88c7087755823
Files :

• pom.xml