Author Archives: nedshawa

About nedshawa

Data wrangler and a Solutions Engineer @Hortonworks

Apache Zeppelin Walk Through with Hortonworks

Screenshot 2015-10-27 11.54.33


Apache Zeppelin (Incubator at the time of writing this post) is one of my favourite tools that I try to position and present to anyone interested in Analytics, Its 100% open source with an intelligent international team behind it in Korea (NFLABS)(Moving to San Francisco soon),  its mainly based on interpreter concept that allows any language/data-processing-backend to be plugged into Apache Zeppelin.

Very similar to IPython/Jupyter except that the UI is probably more appealing and the amount of interpreters supported are richer, at the time of writing this Blog Zeppelin supported:

with this rich set of interpreters provided, it makes on boarding platforms like Apache Hadoop or Data Lake concepts much easier where data is sitting and consolidated somewhere and different organizational units with different skill sets needs to access the data and perform their day to day duties on it as data discovery, queries, data modelling, data streaming and finally Data Science using Apache Spark.

Apache Zeppelin Overview

With the notebook style editor and the ability to save notebooks on the fly, you can end up with some really cool notebooks, whether you are a data engineer, data scientist or a BI specialist.

Zeppelin Notebook Example

Dataset showing the Health Expenditure of the Australian Government over time by state.

Zeppelin also got a basic clean visualization views integrated with it, it also gives you control over what do you want to include in your graph by dragging and dropping fields in your visualization as below:

Zeppelin Drag and Drop
The sum of government budget healthcare expenditure in Australia by State

Also when you are done with your awesome notebook story, you can easily create a report out of it and either print it or send it out.

Car Accidents Fatalities in Melbourne

Car Accident Fatalities related to Alcohol driving , showing the most fatal days on the streets and the most fatal car accident types during Alcohol times

Playing with Zeppelin

If you have never played with Zeppelin before then visit this link for a quick way  to start working it out using the latest Hortonworks tutorial we are including Zeppelin as part of HDP as a technical preview, which may supporting it officially may follow, check it out  Here try out the different interpreters and how it interacts with Hadoop.

Zeppelin Hub

I was recently given access to the beta version of Hub, Hub is supposed to make life in organizations easier when it comes to sharing notebooks between different departments or pepole within the organization.

Lets assume an Organization got Marketing, BI and Data Science practices, the three departments overlaps with each other when it comes to the datasets being used, therfore there is no need anymore for each department to work completely isolated from the others, as they can share their experience together, brag about their notebooks, work together on the same notebook when trying to work on either complicated notebook or different skills are required.

Zeppelin Hub UI

Zeppelin Hub UI

Lets have a deeper look at Hub…

Hub Instances

Instance is backed by a Zeppelin installation somewhere (server,laptop,hadoop..etc), every time you create a new Instance a new Token is generated, this token should be added in your local Zeppelin installation under folder /incubator_zeppelin/conf/ e.g.

export ZEPPELINHUB_API_TOKEN="f41d1a2b-98f8-XXXX-2575b9b189"

Once the token is added, you will be able to see the notebooks online whenever you connect to Hub (

Hub Spaces

once an instance is added, you will be able to see all the notebook for each instance, and since every space is actually either a dept. or a category of notebooks that needs to be shared across certain people, you can easily drag and drop notebooks into spaces making them shared across this specific space.

Adding a Notebook to a Space

Adding a Notebook to a Space

Showing a Notebook inside Zeppelin Hub

Showing a Notebook inside Zeppelin Hub

Very cool !

Since its beta, there is still much of work to be done like executing notebooks from Hub directly, resizing and formatting and some other minor issues, I am sure the All Stars team @nflabs will make it happen very soon as they always did.

if you are interested in playing with Beta, you may request access on Apache Zeppelin website here

Hortonworks and Apache Zeppelin

Hortonworks is heavily adopting Apache Zeppelin, that showed in the contribution they have made into the product and into Apache Ambari, Ali Bajwa one of Rockstars at Hortonworks created an Apache Zeppelin View on Ambari, which gives Zeppelin authentication and allows users to have a single pane of glass when it comes to uploading datasets using HDFS view on Apache Ambari Views and other operational needs.

Apache Ambari with Zeppelin View Integration

Apache Ambari with Zeppelin View Integration

Screenshot 2015-10-27 11.25.15

Apache Zeppelin Notebook editor from Apache Ambari

If you want to integrate Zeppelin in Ambari with Apache Spark as well, just easily follow the steps  on this link


Project Helium is a revolutionary change in Zeppelin, Helium allows you to integrate almost any standard html, css, javascript as a visualization or a view inside Zeppelin.

Helium Application would consists of an View, Algortihm and an Access to the resource, you can get more information of Helium here

Using Apache Nifi to Stream Live Twitter Feeds to Hadoop



with data sources are producing more data over time,with big data evolvement and IoT, with the continuous new sources that enterprises are trying to capture, a mechanism of architecting, visualising the data flow, monitoring and watching the noise that would become signal and a direction for a decision next day, along with enterprise requirements for security like encryption, and data quality at read rather than post process that requires an extensive amount of time from resources.

Apache Nifi (Acquired recently by Hortonworks) comes along with  a web based data flow management and transformation tool, with unique features like configurable back pressure, configurable latency vs. throughput, that allows Nifi to tolerate fails in network, disks, software crashes or just human mistakes…

a full description and user guide can be found here

in this example, we will show how to build an easy twitter stream data flow that will collect tweets that mentions the word “hadoop” over time and push these tweets into json file in HDFS.


  • Hortonworks Sandbox
  • 8GB RAM memory and preferably 4 processor cores.
  • Twitter dev account with Oauth token, follow steps here to set it up if you done have one.


once you have downloaded and started the Hortonworks Sandbox, you can proceed with ssh connectivity to the sandbox, once you are in you would need to download Apache Nifi using the following command:

cd /hadoop
tar -xvzf nifi-0.2.1-bin.tar.gz

after Apache Nifi is now extracted, we need to change the http web port no. from 8080 to 8089 so it doesn’t conflict with Ambari, you can do this by editing the file under /hadoop/nifi-0.2.1/conf/


Confirm installation is successful by starting Apache Nifi and logging to the webserver

/hadoop/nifi-0.2.1/bin/ start

now point your browser to the IP address of the Sandbox followed by the port no. 8089, in my setup it looks like the following:

Creating a Data Flow

In order to connect to Twitter, we would either create a whole data workflow from scratch, or use a twitter template that is already available under the templates from Apache Nifi here, from these templates we will download the pull_from_twitter_garden_hose.xml file and place it on your computer.


once the template is downloaded we can go to the webserver and add the template by clicking the template button on the left side corner and browse for the downloaded template to add it.

Template Uploading

now Browse to the xml file downloaded in the previous step and upload it here.

Template Confirmationyou will see the template that you have installed as marked in red, once this step is completed, lets add the template to the workspace and start configuring the processors, you can do that by adding a template using the button on the right top corner as followingAdd TemplateChoose Template

once you add the template you will end up with something like this:

Twitter Templateyou can start discovering every processor, but in a nutshell this is what every processor is doing:

Grab Garden Hose: Connecting To twitter and downloading the tweets based on the search terms provided using the twitter stream api.

Pull Key Attributes: evaluates one or more expressions against the set values, in our case its SCREEN_NAME,TEXT,LANGAUGE,USERID and we want to make sure the value is not NULL for those otherwise it will not mean much to us, the criteria is set to on “Matched” which means only Matched criteria will be passed to the next processor.

Find only Tweets: filters tweets that has no body message, retweets..etc

copy of Tweets: this is an output port that copy these into the Apache Nifi folder under /$NIFI_HOME/content_repository

now, lets create a new processor that copies the data to HDFS, but before we do that lets create the deistination folder on Hadoop by using the following command

hadoop fs -mkdir -p /nifi/tweets

now lets add the processor by clicking on the processor button on the top right corner

Add Processor

Now, lets choose the PutHDFS Processor, you can easily search for it in the search bar on the top.

putHDFS Processor

connect the HDFS Processor to the Find Only Tweet Processor and choose “tweet” as a relationship

Screenshot 2015-09-02 10.56.28
Screenshot 2015-09-02 10.19.33
now right click on the putHDFS processor and configure, you need to choose how the processor will terminate the flow, and  since this is the last processor in the flow and it wont pass any data beyond, tick all the auto-termination boxes.

Screenshot 2015-09-02 10.59.49

and go to the properties tab and add the hdfs-site.xml and core-site.xml locations, usually they are under /etc/hadoop/ if you are downloading the latest 2.3 sandbox/ , also dont forget to add the folder that we have created earlier on HDFS.

Screenshot 2015-09-02 11.00.51

hopefully after this you should get a red stop button instead of a warning logo on the processor box, if not check what the error is by keeping the cursor on the warning logo.

lets configure the twitter processor and add the consumer key and the access token information, notice that the consumer secret and the token secret are encrypted and wont be visible, also make sure you change the twitter end point to “Filter Endpoint” otherwise your search terms wouldn’t be active.

Screenshot 2015-09-02 10.12.52

Screenshot 2015-09-03 08.15.40

as you can see in the search terms we have added the word “hadoop”, once you are done, verify with the warning logo disappearing from the box.

now we are ready to start the flow, just simply right-click on each box and start, you should start seeing numbers and back pressure information, after a while you can stop it and verify that json files are now stored on HDFS.

Screenshot 2015-09-02 10.30.09


you can always add more processors to encrypt, compress or transform the files to Avro or SEQ files for example prior to dropping them in HDFS, a complete list of the available processors can be found here

Connecting to Hive Thrift Server on Hortonworks using Squirrell Sql Client



Hive is one of the most common used databases on Hadoop, users of Hive are doubling  per year due to the amazing enhancements and the addition of Tez and Spark that enabled Hive to by pass the MR era to a an in-memory execution that changed how people are using Hive.

in this blog post, I will show you how to connect squirrel Sql Client to Hive, the concept is similar to any other clients out there as long as you are using the open-source libraries that matches the ones here you should be fine.


Download Hortonworks Sandbox with HDP 2.2.4Squirrel SQL Client

Step 1

Follow the Squirrel documentation and run it on your Mac or PC.

Step 2

Follow the Hortonworks HDP Installation on VritualBox, VMware or Hyper-V and start up the virtual Instance.

Step 3

once you are HDP is up and running, connect it it using SSH as it shows on the console, once you are connected you need to download some JAR files in order to establish the connection.

Step 4

if you are using MacOS, simply while you are connected to you HDP instance search for the following JARs using the command:

root> find / -name JAR_FILE

once you find the file needed, easily copy it using SCP to your laptop/PC

root> scp JAR_FILE yourMacUser@yourIPAddress:/PATH_TO_JARS

the files you should look for are the following (versions will differ base on which Sandbox you are running but different versions are unlikely to cause a problem)

  • commons-logging-1.1.3.jar
  • hive-exec-
  • hive-jdbc-
  • hive-service-
  • httpclient-4.2.5.jar
  • httpcore-4.2.5.jar
  • libthrift-0.9.0.jar
  • slf4j-api-1.7.5.jar
  • slf4j-log4j12-1.7.5.jar
  • hadoop-common-

if you are running windows you might need to install winSCP in order to grab the files from their locations.

Step 5

Once all Jars above are downloaded into your local machine, Open up Squirrell and go to Drivers and Add New Driver.

JDBC Driver Configuration for Hive

Name: Hive Driver (could be anything else you want)
Example URL: jdbc:hive2://localhost:10000/default
Class Name: org.apache.hive.jdbc.HiveDriver
go to Extra Class Paths and add all the JARS you downloaded

you may change the port no or IP addresses if you are not running with the defaults.

Step 6

login to you Hadoop Sandbox and verify that HIVESERVER2 is running using:

netstat -anp | grep 10000

if there was nothing running you can hiveserver2 manually

hive> hiveserver2
Step 7

once you verify hiveserver2 is up and running you are ready to test the connection on Squirrel by creating a new Alias as following

Alias Creation for JDBC connection

you are now ready to connect, once connection is successful you should get a screen like this

Connection Established Screen

Step 8 (Optional)

With your first Hive Query, Squirrel can be buggy and complain about memory and heap size, if this ever occurred, if you are on Mac, right click on the app icon–>show package contents–>open info.plist and add the following snippet


Now you can enjoy…

Hadoop+Strata San Jose 2015 Highlights


Being at Hadoop Strata 2015 at San Jose is like being transported to the future and see how the software technology will look like less than a decade from now.

I have visited Hadoop Strata all the way from Melbourne – Australia to try to catch up with all the new vendors in the market and bring this expertise to the rest of world as many of these products are focusing in their region.

Now lets get into business,…..

Apache Spark


Apache Spark is still looking like the way to go when it comes to the fading map reduce andApache Mahout being transitioned to spark MLLib, we had a lack of presistant tables storage for databases in spark, we were able to get away with registering temporary tables but now all changed for Spark with the announced BlinkDB that will be holding the metastore and tables storage.

spark_roadmap.jpgR on spark is a very aggressive move as it eliminates the need to program R outside of Spark eco-system, allowing analytics within the eco system using the same RDDs (Resilient Distributed Data sets) that are living in memory.

the main benefits of the spark platform is the ability to stream data using Spark Streaming, query data using SparkSQL (previously Shark) and apply analytics using MLLib and and R within the same platform utilizing the same RDD that is occupying the memory, therefore boosting the performance since the data is moved to memory at the beginning eliminating any need to read from spinning disks as long as their is a sufficient memroy to hadnle the data or even partially in memory as Spark by nature spills over to Disk when there is insufficient memory to keep all the datasets in.

the question that I get asked all the time, can spark used independently from Hadoop? and the answer is Yes, Spark can run in a standalone mode or using a cluster like Apache Mesos, the only problem is Spark will become another silo rather than being integrated in Hadoop and utilizing the different data stored from the mixed workloads and the Data Lake.

Unified Data Storage on Hadoop


As part of setting standards in working with Hadoop files, there is a big move and efforts to push for data standardization using Avro and Parquet in order to avoid the chaos of the different files format and extensions.

Avro is used for row based storage format for Hadoop and Parquet is used for columnar based storage, depending on how the data will be dealt with and what type of a file its we can easily decide how to store it, for most of the time Avro is the way to go unless we are working with a dataset that consist of plenty of columns and I want to avoid scanning un-needed columns as they wont be used much, the I can easily select a Parquet format and pre-select my columns in order to optimize my queries for examples in Hadoop.

Cloudera was showing the Kite SDK that can easily convert the data into Avro or Parquet format, we can also integrate Kite to do the conversion at the point of ingestion if needed. Most of the current databases in Hadoop can read Avro and Parquet formats (Pivotal Hawq, Cloudera Impala, Hive, Spark…etc)

Hive on Tez vs Hive on Spark

hive_logo.pngthe future of relational databases is here now, from my point of view its only a matter of time until companies starts replacing their existing traditional ones to the Hadoop ones like Pivotal Hawq or Apache Hive…etc. I have attended a smart talk by a smart performance engineer Mostafa Mokhtar from Hortonworks, they were trying to benchmark every aspect of Hive stand alone compared to Hive on Tez and Hive on Spark, we were hearing a lot of noise on how disruptive Hive on Spark would be, surprisingly Hortonworks was able to show that Hive on Tez is more than 70% faster than Hive on Spark, although Spark-SQL is more than %60 faster than hive on Spark and probably faster than Hive on Tez as well.

this is a great news for current Hive users as its obvious that Hive on Tez is the way to go on this one.


The Rise of Analytics, Hadoop and The Data Lake – Analytics

In my current role, I am being so lucky getting insights from different customers around the rise of their analytics story, and the baby steps companies out there are taking in order to get faster insights on their customer base before some competitor does in order to retain the customer base and possibly grow it, along with the baby steps some giant mistakes are occurring returning the company a thousand step backwards, especially in a field where FINALLY application and storage meet together outside performance requirement and IOPS discussions but talking about the governance of the data, storage , access, protection and most importantly Automation.

In the last 2 decades of data storage and with more and more application being developed in organizations, data ended up in silos of storage, some are still being tier one served from a tier one SAN storage other are shared folders and web logs being stored in a slower more economic and denser tier which could be a NAS storage.

The last big step in the database world was data warehousing, where many of databases where living isolated from each other, not allowing for a collaborative way of achieving analytics efficiently in an organization, over the last decade most of the CIOs task list was about the Data Warehousing Strategy.

Analytics on databases is a way of creating complex (sometime not) queries or searches from a data source(s), others may call it data mining well all , in my opinion, There are three different stages for analytics.

Stage 1 started with single database analytics, Stage 2 started a decade ago with analytics performed on multiple data sources and databases mostly using data warehousing where the data was mostly structured, Stage 3 that is discovering or just about to implement it which is analytics based on the same old structured data that we use to query in addition to the unstructured data.

Handling structured data is so mature now that most people understands that I need a relational database to handle my customer orders records for example (Although NOSQL databases are a valid replacement for this market), unstructured data came in with its own challenges, I can write a an 3000 word article on each of them but shortly for this one its data ingestion and flow where tools like Apache storm and Pivotal springXD excels at, software storage layer where a file system like Hadoop is the preferred way to go depends on the case you may use MongoDB, Cassandra…etc a friend of mine actually forwarded me a great article on when to use what

Unstructured data varies from machine generated data (e.g. server logs) web logs (cookies, customers movement on the web, tracking..etc) social media like twitter, facebook, mobile phones beacons (Bluetooth and Wifi) and any other information that relates to your business, customers or product.

These days organizations are taking analytics really seriously, I can see these job ads everywhere for data engineers and data scientists (yes there is a major different), I have met tons of these data scientists here and no not all of them are R experts, some of them only know couple of visualization tools like Tableau but they are able to get great insights from the data, some of them doesn’t even know basic shell commands on unix, and some of them are completely Microsoft windows only users….

Retail, City Councils, Airports, Insurance, Finance, Banking, Government, Defense, and even small startups have started doing something around analytics, infact some of them talk about innovation and Gartner Pace Layered approach (Open Data Lake), in Airports we talk about flight delays prediction, Bluetooth beacons, city councils does parking sensors on the streets for sending the inspector when time is expired for a certain occupied bay and traffic light management using beacons, Retail needs better insights and not the overnight ones from EDW,also their massively increasing foot print of data, insurance collects data from mobile apps in order and correlate customers to their information in order to retain their footprint, health insurance are giving away fit bits to offer better discounts for active healthy customers, cant talk about government and defense.

in the next episodes I will be  connecting more dots together in technology prespecitves, hadoop, in-memory databases, mpp databases,reporting tools,lambda architecture,gartner stuff, …etc stay tuned…

Twitter Analysis for #Emcworld 2014 using SpringXD, Pivotal, Hawq, Isilon and Tableau


Did you ever wonder what people were mainly talking about in #EMCWORLD? what are the most products that were grabbing most of the attention of attendees? who were the top Twitter contributors? how many Tweets did we get for #EMCWORLD this year?

#EMCWORLD 2014 Twitter Analysis Results


EMC World ran from 5th of May to the 8th of May 2014 in Las Vegas, I was able to collect around 30,000 tweets, unfortunately there were many missing tweets due to the twitter API limitation at this point and not being able to consistently get all updates as Twitter restricts you by the rate of data being pulled, but at least I am covering no less than 70% of the twitter data in that period which is usually enough to make an idea of the top stuff that was going on.

I have done my best efforts to make this simple, however due to limited time to put this together they might be some missing pieces or not in depth explained steps that I will be refining over the next couple of weeks, therefore I appreciate your patience.

I have used many products in order to get this going, make sure you read the terms and conditions before downloading and agree on the terms on your own risk (although most of them are free for at least 30 days trial, and the rest is free for personal/educational purposes).

I have not covered Isilon Storage Nodes piece in this article, however I will be updating it shortly to show you how to connect isilon into PHD 1.1 so please stay tuned.


Running analysis on Twitter is one emerging way of analysing human responses to products, politics and events in general, for many organization out there understanding the market behaviour is through what people mention on Twitter, who are the most people that tweet with the most influence…etc

The outcome of such analysis is usually GOLD to marketing departments and support organization that can make sure customer satisfaction is guaranteed by capturing outcome of marketing events,surveys, product support, general product satisfaction..etc .etc

The Hadoop Part

In this exercise all tweets from a specific hash-tag will be streamed to Hadoop, in my case I am using Pivotal HD and storing the data on HDFS Isilon Storage (not local disk), usually having a Scale Out HDFS storage like Isilon is an ease of mind for enterprises as you separate your compute layer from the storage layer, better protection of data with tiering of disk pools, adding more compute with no storage, and you can still serve file-server  / Home Directories / and lots of more workload on the same storage, for more information on why Isilon and Hadoop click here.


Another piece of the puzzle is SpringXD, it allows you to connect many sources as Mail,Http,Twitter,Syslogs,Splunk….etc it has ready made APIs to connect to sources and grab you the information you are after, then get sinked to sink of your choice e.g. HDFS,FILE,JDBC,Splunk..etc, in our case the source is Twitter and the  Sink is HDFS for more information on how to connect different sources and sinks click here.


Once SpringXD is up and Running we need a way to query data, and that is why we are using Pivotal HAWQ which is the fastest SQL-Like implementation on Hadoop until now, Hawq will enable us to create external tables through PXF on the fly, once we are done we can create views and manipulate the data in SQL if we want to and export it later on.


Tableau will come in place after all data is gathered so we can create a pretty visualization of a mixture of text data coming out from SpringXD and table data coming out of Hawq. you can always use different software like SpotFire from TIBCO or Cognos.

Now lets start

Prep Work

– Download PHD VM 1.1 (currently its up to 2.0 but unfortunately SpringXD only supports PHD 1.1 at the moment same applies to Isilon) from here
– Download SpringXD M6 from here
– Create a Twitter account if you dont have one
– Sing-in to, create an application and generate Keys, take note of all the keys.

– Download WINSCP and transfer SpringXD to your Pivotal HD host from here


Deploying Pivotal HD

This is a straight forward process when you just fire the VM up with Vmware workstation and it should properly work, make sure you assign enough Memory and CPU by default it will take 4GB of RAM and 1 CPU,1 Core, I would bring it up to 8GB and 2 CPU or at least 2 cores if possible.

btw, credentials are usually root/password and gpadmin/Gpadmin1

Deploying SpringXD

Once you have PHD 1.1 running, using root user move the springXD folder to /home/gpadmin then start the sequence of the following commands:
using root:

#chown gpadmin:hadoop /home/gpadmin/<path to springXD>
#chmod 770 /home/gpadmin/<path yo SpringXD>

using gpadmin:
#unzip <home/gpadmin/<path to sprinXD>

#vi ~/.bashrc
once you are in the file using vi add the following row
export XD_HOME=<path to SpringXD>
export JAVA_HOME=<path to JDK directory>

springxd export

We now need to start our HDFS platform and SpringXD (using gpadmin):



#vi /<path-to-xd>/xd/config/servers.yml

uncomment all the lines from spring to fsUri

#~/<path to SpringXD>/xd/bin/start-singlenode –hadoopDistro phd1


Twitter’s API

Twitter uses two APIs for communication, TwitterSearch and TwitterStream, you can read in depth on the differences here but in my case TwitterStream kept coming back with rate-limit issues due to the amount of tweets I was getting, but I will still show you how to create both of them but before that lets check that everything is ready for SpringXD.

Reformatting Tweets with Python

using gpadmin, #hadoop fs -ls / , at this point you should see something like this:

if you cant see it, then PHD was not started, so start it up first, after that lets work on a little script that I have actually used from the guys here, this script that you can download from here will allow us to reformat the tweets from json, however we will need to modify it a little bit to get the text of the tweets as the script below

import groovy.json.JsonSlurper

def slurper = new JsonSlurper()
def jsonPayload = slurper.parseText(payload)
def fromUser = jsonPayload?.fromUser
def hashTags = jsonPayload?.entities?.hashTags
def followers = jsonPayload?.user?.followersCount
def createdAt = jsonPayload?.createdAt
def languageCode = jsonPayload?.languageCode
def retweetCount = jsonPayload?.retweetCount
def retweet = jsonPayload?.retweet
def text1 = jsonPayload?.text
def id = jsonPayload?.id

def result = “”
if (hashTags == null || hashTags.size() == 0) {
result = result + + ‘\t’ + fromUser + ‘\t’ + createdAt + ‘\t’ + ‘-‘ + ‘\t’ + followers + ‘\t’ + text1 + ‘\t’ + retweetCount + ‘\t’ + retweet
} else {
hashTags.each { tag ->
if (result.size() > 0) {
result = result + “\n”
result = result + + ‘\t’ + fromUser + ‘\t’ + createdAt + ‘\t’ + tag.text.replace(‘\r’, ‘ ‘).replace(‘\n’, ‘ ‘).replace(‘\t’, ‘ ‘) + ‘\t’ + followers + ‘\t’ + text1 + ‘\t’ + retweetCount + ‘\t’ + retweet

return result

save this script under ~/<path to springXD>/xd/modules/processors/scripts/reformat.script

Creating SpringXD Streams

Now we are ready to create our first stream, lets start with TwitterSearch, its usually easier and get the job done straight forward, remember the API keys are masked in this example therefore make sure you replace them with the one you have, as for this example we will query a hashtag let it be #Australia, which means we will be able to get all tweets that contains the #Australia hashtag, make sure you set the delay for at least 10000 otherwise you will get an error of rate-limit very soon:

using gpadmin, run to ~/<path-to-springxd>/shell/bin/xd-shell

xd-shell> hadoop config fs –namenode local


xd-shell> stream create –name Emc–definition “twittersearch –consumerKey=xxxxxoxRYkIQZzDgMicxxxx –consumerSecret=bZvxxxxx51hld3UXObHlCy4rc6bmXJhEMH3TjGIndDxxxxxN –query=’#Australia –fixedDelay=10000 –outputType=application/json | transform –script=reformat.script | hdfs –rollover=1000” –deploy

Now lets create a stream using the TwitterStreamAPI in addition to the search one that we have just did:

xd-shell> stream create –name TestTwitterStream –definition “twitterstream –consumerKey=xxxxxoxRYkIQZzDgMicxxxx –consumerSecret=bZvxxxxx51hld3UXObHlCy4rc6bmXJhEMH3TjGIndDxxxxxN –accessToken=xxxxx71851-pf9IWlpWiKT9mgk84pR9K6GFOrZiyWxxxxxxxxx –accessTokenSecret=pgQ2Yxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx –track=’#Emc’ | hdfs –rollover=1000” –deploy


by the way, SpringXD is supported to run on windows, so you are welcome to run it from an external client as long as you configure the IP address or the dns name when you start xd-shell using the command:


hadoop config fs –namenode hdfs://<hdfs-IP-ADDRESS-OR-DNS>:8020

Validating the Data

Once we have all tweets required for the analysis, lets say you were gathering tweets for hours, days…etc on a specific subject (In my case I was tracking #Emcworld event in Las Vegas for 4 days)

lets see how my files are looking like from the shell by using the following command with gpadmin user:

#hadoop fs -ls /xd/<the folder name>

hadoop files

you can calculate the sizes using the following command

#hadoop fs -du sh /xd/

Connecting to HAWQ

Now, all data looking good, I need to create my tweets table using the information I have so I can do some awesome queries and fetch some results, I have used the same command used by some pivotal experts like here

id BIGINT, from_user VARCHAR(255), created_at TIMESTAMP, hash_tag VARCHAR(255),
followers INTEGER, tweet_text VARCHAR(255), retweet_count INTEGER, retweet BOOLEAN)
LOCATION (‘pxf://pivhdsne:50070/xd/<directory of your tweet texts>*.txt?Fragmenter=HdfsDataFragmenter&Accessor=TextFileAccessor&Resolver=TextResolver’)

The output will show you a table that consists of all the HDFS files ingested into a the HAWQ table, from there you can start querying the table and get the information you are after, sometimes and depending on how big the files are you might need to clean up the files, mainly with another python script or manually using excel but this means you have to export the data merge them and reformat them with excel then back to the cluster once you are done, most of the clean up tasks will be around deleting null or empty values, you might wanna remove duplicates if any or re-sort the tweets.

Now, lets look on how Hawq is querying the data, lets for example see how many tweets do I have (to enter the psql prompt, jst type psql)

lets see what is the total tweets that we have received during EMCWORLD




Now lets see who was the top tweeters


I was wondering how many retweets we had vs original tweets







Now, I wanted to know how many mentions we had for most of EMC products, so I have created a table including all the products, and then created a view joining the tweets table the and product table to get the mentions per product code below:

create view products_tweets as select product, count(a.*) product_count from tweets a,emc_products b where a.tweet_text like ‘%’||||’%’ group by order by count(a.*);

The result was fascinating

Exporting Data to Tableau

After we had all the fun with SQL commands and queries, I wanted to have a proper visualisation of the results, I had exported the files and the output of view table to text file and ingest it into Tableau, started playing with Tableau and that was the outcome.

If you are new to Tableau, you can go through some free tutorials and examples that will give you an idea how to start, usually its pretty simple unless you started to have more complex models, then a proper training is usually recommended.


below is one of the working sheets in Tableau showing the Top Hashtags being used without running a sql query.Tableau Results

Top Hashtags

Remember to comment or like my post if you enjoyed it, please feel free to ask any question, I am regularly updating this post to make sure there are not issues with downloading and I am refining the commands regularly to make it easier to deploy such analysis.


PACS Systems and Scale out NAS

Today I had an interesting meeting with an intelligent engineer from a major PACS provider discussing how the technology is evolving and how a true scale-out NAS can add value and reduce the amount of work needed and providing a better use experience when it comes to PACS.

I have personally worked around HIS,RIS and Imaging systems for a long time, only knowing the basics, I was responsible of the designing and implementing the infrastructure for over 20 Hospitals that will eventually host HIS,LAB,RIS,PACS…etc .

our discussion today was around  the challenges, how it integrates with RIS…etc etc I started the conversation with “How can a storage help?”, the first answer he came up with was “By avoiding PACS migrations!!” Then we started talking about the pains in PACS migrations, and what are the problems everyone faces, before I get to migrations, let me take you through PACS and the components around it:


PACS stands for Picture Archiving and Communication System, from the name you can tell we are manging pictures mainly, the source of these pictures can be a variety of imaging machines like CT Scan, MRI, X-Ray..etc, PACS organizes the pictures into a specific order and connects them to RIS via certain keys (Patient No,, Visit,..etc), although you can run PACS as a standalone system for small practices where RIS is not necessary.

PACS uses DICOM as a standard for transmitting, receiving and storing the images. PACS usually is a application/database software runs a normal transactional database (e.g. Oracle/SQL..etc)  most of the times PACS runs a standalone database to store DICOM images headers.

PACS has many vendors in the market such as GE, AGFA, FUJI, KODAK, MCKESSON, ACUO…etc Many vendors are now moving to virtualizing PACS by providing it in a Virtual Appliance, to easy deployment and reduce costs, however the integration part with RIS/HIS is still manual due to the different HIS/RIS suppliers out there and different communication methods.


RIS (Radiology Information System) usually takes care of scheduling machines availability and appointments, results and findings store, billing (specially if its a stand alone RIS/PACS with no HIS), and further features can be derived from the HIS (Hospital Information System) if available and integrated with RIS. Integration with RIS is through another standard called HL7.


Digital Imaging and Communication in Medicine, which is mainly a standard to handle all different operations from transmitting,recieving,storing images with its own file format, its a standard developed by members of NEMA (National Electrical Manufacturers Association) and it currently holds the copyrights for DICOM.

How does it work?

Once a Modality ( Medical term referring to imaging devices) operator starts the imaging procedure, The modality will start transmitting the images according to the PACS system, the PACS system responds with error messages if any failure occurred during the transmission phase. once all the images are received in the PACS, it starts creating headers (Metadata) for the images, the header will usually be derived from RIS/HIS or manually if RIS/HIS doesn’t exist, the header is written according to DICOM format, and usually contain TAGS, which are usually numbers that translates into a standard key attributes (e.g. 008,0070 refers to Modailty Manufacturer) same for patient ID, Visit and over another 100 attribute (mostly fetched from HIS/RIS), eventually all the attributes are stored on the PACS database.


So why do we need to migrate our PACS data?

There are many reason to migrate PACS like vendor change, VNA, Archiving…etc which will all take us back to the same basic length element in migrations:

Storage LimitationMost of the non-scale out storage are usually a victim, once storage maximum capacity is reached or its time for a new storage or you are migrating archival PACS to a separate storage, or the new vendor suggested another storage to work with or changing protocols, eventually , you have to take the old storage out and get the new one, but guess what!! you have to migrate all data to the new storage first 🙂

PACS Migration

imagine an MRI imaging device producing 400 image approximately per patient, each image is around 4MB, its booked for 1 hour per visit, and you can have 8 patients a day, lets say the imaging center is moving to a new storage after 3 years, so according to simple math we have to worry about around 10TB of data, Now start add CT scanning (around 100 images), and X-RAY and multiply by the no. of the machines. its not rare to see Petabytes at some hospitals, depending on how long the images are retained and the compliance regulations around retention the data stored can get really massive.

Migrating data takes two approaches, the first is the easy one which is mainly images with no headers, usually this is a quick and easy it depends on how big is your data and how fast you can transmit them (what media you are using). The other approach is the dirty one, which uses the vendor migration tools and can takes several months to years, especially when moving through vendors, each image header is inspected, database is queried, new database record in the new system is inserted, an complete pre-assessment engagement usually takes  a place first to check the integrity of the files and headers and after cleanup and assessment the migration can start in phases, with a complete assessment after each phase is completed with logs revision…etc

Common Issues with PACS Migration

One of the most problems that affects data integrity is the header (Metadata) of the images, lets say the patient ID was 1234, and there was no HIS/RIS integration and the operator manually added the wrong patient ID at the time of the scan 1235, once the images are transmitted to PACS. The header of the image will have 1234 as a Patient ID as well as the PACS , based on DICOM, some tags can never change, therefore even if you update the PACS with the new patient ID only a  reference with the new patient will be written in the PACS reference the wrong patient ID in the header so 1234 in PACS and 1235 in the header, but PACS creates a link. such problems are fixed during the pre-assessment stage and therefore it may be time consuming and costly.

Choosing the right storage

Scale-out NAS storage over traditional storage will eliminate the need for storage migrations due to the following reasons:

– Can expand on the fly with no interruption

– Can be upgraded on the fly with no interruption

– Can have slower nodes with cheaper disk for archiving on the fly using file policies to move the images to slower disk without affecting the path for the files (so database can still reference the images in the same path)