Building node-accumulo for Accumulo 1.5

[cmundi]

Thanks for putting your code on github. I'm learning from it! In fact I want to access accumulo from node with a message queue and (boom!) Google gave me your repo. Awesome.

Unfortunately, the java throws an exception, down in Accumulo's implementation of ZookeeperInstance calling on Thrift:

Exception in thread "main" org.apache.accumulo.core.client.AccumuloException: org.apache.thrift.TApplicationException: Internal error processing authenticateUser

The exception occurs during setup in AmqpWebAnalytics.java on this line of code:

this.connector = instance.getConnector(this.username, this.password.getBytes());

Execution never gets around to testing for table existence.

The issue appears to be that getConnector(String user, ByteBuffer pass) is now deprecated in ZooKeeperInstance: http://accumulo.apache.org/1.5/apidocs/org/apache/accumulo/core/client/ZooKeeperInstance.html

---

Here's my setup:

Accumulo 1.5 ... instance = a01, user=root, password=password. (Really!) Zookeeper 3.4.5 Hadoop 1.2 node 0.10.15

The Accumulo shell works great, so that'snot the problem. I set the accumulo instance name, user and password to match my setup, and then I built the java app.

Any ideas? I'm sure I'll figure this out but maybe this info will help someone else in the meantime.

Thanks!

[joshelser]

Ha, small world.

Yeah, I wrote this before 1.5 was released with the new API calls. You could try updating it to use the new PasswordToken construct (instead of the getBytes on the password String). I would have expected this to still work against 1.5, but there may be something unexpected happening with Accumulo itself.

I'll try to bring this up to date against 1.5.

[cmundi]

Hi Josh. I've done exactly that -- adopted the new PasswordToken and updated the imports. Of course I also had to update the pox.xml to reference the updated versions. Then the fun starts...the compile fails with this:

[ERROR] Failed to execute goal on project node-rabbitmq-webanalytics: Could not resolve dependencies for project accumulo:node-rabbitmq-webanalytics:jar:0.0.1-SNAPSHOT: The following artifacts could not be resolved: javax.jms:jms:jar:1.1, com.sun.jdmk:jmxtools:jar:1.2.1, com.sun.jmx:jmxri:jar:1.2.1: Could not transfer artifact javax.jms:jms:jar:1.1 from/to java.net (https://maven-repository.dev.java.net/nonav/repository): No connector available to access repository java.net (https://maven-repository.dev.java.net/nonav/repository) of type legacy using the available factories WagonRepositoryConnectorFactory -> [Help 1]

The javax.jms complaint popped up as soon as I required Accumulo 1.5 in the pom.xml. I' haven't battled with jms in a long time, so I might have to sleep on this one...

Thanks!

P.S. Yes, the world is small. It's a large graph with small max{edges(e1,e2)}. ;)

[cmundi]

FYI. Updating rabbitmq to 3.1.3 neither helps nor hurts as far as I see so far. I was hoping it might pull in a substitute for jms which does not require JEE. Nope.

<version.amqpclient>3.1.3</version.amqpclient>

UPDATE 1: I'm making progress by excluding the JMS stuff from the ZooKeeper dependency in pom.xml.

UPDATE 2: Also exclude cloudtrace dependency. Now I can compile, but I get another exception:

Exception in thread "main" java.lang.NoClassDefFoundError: org/apache/accumulo/fate/zookeeper/ZooCache
    at accumulo.AmqpWebAnalytics.setupAccumulo(AmqpWebAnalytics.java:87)

At this point, I am stuck. The docs for Accumulo 1.5 indicate that o.a.a.fate.zookeeper.ZooCache does exist. So I wonder if this might be an internal issue in Accumulo.

[cmundi]

Ok, so I finally just listened to the diagnostic. For whatever reason, maven did not copy accumulo-fate-1.5.0.jar to the webanalytics/target/lib. I dropped in the jar I built, and I got past the ZooCache. Now the classloader says I'm missing org/apache/commons/configuration/Configuration. Let's see how deep this rabbet(mq) hole goes...

[joshelser]

Depending on the version of Hadoop used, you may be missing dependencies that Accumulo marked a provided (or that I marked as provided).

commons-collections, commons-configuration, commons-lang and commons-io come to mind.

[cmundi]

Aha! We also need apache-commons and apache-lang. I think I've got it now. I will fork and prepare a branch for consideration.

Thanks!

UPDATE: Ooops. Still not happy with thirft. This one appears to be in flux. I might actually need to figure this maven thing out for real some day.

[cmundi]

Ok, I got it working but it's not pretty. I finally gave up on trying to outsmart maven. I just built libthrift 0.9 from source and dropped the jar in my lib folder. The node-accumulo demo is now working as expected. Yeah!!!

Altogether, accumulo 1.5 API was very simple to accomodate. My problem was raven. I guess I need to break down and buy the anteater book when the new edition comes out next year.

So I ended up with these versions, with corresponding additions further down in the pom:

    <version.accumulo>1.5.0</version.accumulo>
    <version.zookeeper>3.4.5</version.zookeeper>
    <version.gson>2.1</version.gson>
    <version.amqpclient>3.1.3</version.amqpclient>
    <version.hadoop>1.2.0</version.hadoop>
    <version.commonslogging>1.0.4</version.commonslogging>
    <version.log4j>1.2.17</version.log4j>
    <version.thrift>0.9</version.thrift>
    <version.slf4j>1.6.4</version.slf4j>
    <version.commonsio>1.4</version.commonsio>
    <version.commonsconfiguration>1.9</version.commonsconfiguration>
    <version.commonslang>2.6</version.commonslang>

There are three misleading aspects of what appears above. I was not able to get maven to do all my work for me, so I manually copied libthrift-0.9.jar, accumulo-trace-1.5.0.jar, and accumulo-fate-1.5.0.jar into my lib directory. I didn't check which version of the Thrift API is actually needed by Accumulo 1.5, but 0.6.1 does not work and I just built latest and got lucky.

The code changes for the new API are trivial:

+import org.apache.accumulo.core.client.security.tokens.PasswordToken;
.
.
.
+  protected PasswordToken token = null;
.
.
.
-    this.connector = instance.getConnector(this.username, this.password.getBytes());
+    this.token = new PasswordToken(this.password.getBytes());
+    this.connector = instance.getConnector(this.username, this.token);

I would create a pull request if I had better luck with maven. :(