bigdata

Consistent Hashing

abgoswam algorithms, bigdata

References:

 

Advertisements

Mahout Spark Shell Locally

abgoswam bigdata, mahout

I was playing around with Mahout, and one of the things I wanted to try out was to use Mahout’s Spark Shell on my local machine

There is a nice example for doing this. But I hit a stack dump the moment I tried to start up the mahout shell using bin/mahout spark-shell

<br />java.lang.RuntimeException: java.io.InvalidClassException: org.apache.spark.rpc.netty.RequestMessage; local class incompatible: stream classdesc serialVersionUID = -2221986757032131007, local class serialVersionUID = -5447855329526097695
at java.io.ObjectStreamClass.initNonProxy(ObjectStreamClass.java:616)
at java.io.ObjectInputStream.readNonProxyDesc(ObjectInputStream.java:1630)
at java.io.ObjectInputStream.readClassDesc(ObjectInputStream.java:1521)

The problem is because the spark version that Mahout was looking for was 1.6.2  (specified in the POM file).  The spark cluster I had started up was with the latest version 2.0.1

Here are the steps I did to get it going:

Installing Mahout & Spark on your local machine

  •  Create a directory for Mahout somewhere on your machine, change to there and checkout the master branch of Apache Mahout from GitHub :
  • Change to the mahout directory and build mahout using mvn -DskipTests clean install
  • Download Apache Spark (http://www.apache.org/dyn/closer.cgi/spark)
    • Note: Download the source code not just the pre-built binaries.
    • Select ‘Source Code’ in the Project type
  • Change to the directory where you unpacked Spark and type `sbt/sbt assembly` to build it
    • Takes close to an hour

 

Starting Mahout’s Spark shell

  • Goto the directory where you unpacked Spark and type `sbin/start-all.sh` to locally start Spark
  • Open a browser, point it to http://localhost:8080/ to check whether Spark successfully started. Copy the url of the spark master at the top of the page (it starts with spark://)
    • This starts spark in the Standalone mode with 1 master and 1 worker
    • Verified the spark version used was 1.6.2
  • Define the following environment variables in a file `mymahoutsparksettings.sh` and source that file so the following variables are set
<br />abgoswam@abgoswam-ubuntu:~/repos/mahout$ cat mymahoutsparksettings.sh
#!/usr/bin/env bash

export MAHOUT_HOME=/home/abgoswam/repos/mahout
export SPARK_HOME=/home/abgoswam/packages/spark-1.6.2
export MASTER=spark://abgoswam-ubuntu:7077

echo "Set variables for Mahout"
abgoswam@abgoswam-ubuntu:~/repos/mahout$
  • Finally, change to the directory where you unpacked Mahout and type `bin/mahout spark-shell`, you should see the shell starting and get the prompt mahout>.

References:

Windowing Operations in Azure Stream Analytics

abgoswam azure, bigdata, sql

Windowing is a very common operation in stream analytics.

Beneath the surface, there is a whole bunch of complex data structuring that’s going on to support the windowing operations. I would love to dig deeper into these someday.

Example:

Here is an example of a query I wrote recently using windowing operators in azure stream analytics. It shows 3 interesting things :
1. Windowing
2. CTEs
3. Aggregation over string columns (using TopOne)

WITH ContextReward AS (
    SELECT 
        eventid,
        TopOne() OVER (ORDER BY [EventEnqueuedUtcTime] ASC) CR,
        MAX (reward) AS reward
    FROM Input
    GROUP BY eventid, HoppingWindow(Duration(hour, 2), Hop(hour, 1))
)

SELECT 
    reward,
    eventid, 
    CR.actionname AS actionname,
    CR.age AS age,
    CR.gender AS gender,
    CR.weight AS weight,
    CR.actionprobability
INTO OutputWindow
FROM ContextReward

SELECT * INTO Output FROM Input 
SELECT * INTO OutputCSV FROM Input

 

References: