Data In Motion: March 2019

Using Sensors At Scale on the Edge Part 1 : Apache NiFi : MiNiFi : CEM

In the next release I will be adding CEM.

This is an update to an existing flow that adds new features based on Apache NiFi 1.9.1. We will also be migrating this to MiNiFi 0.6.0 C++ with Cloudera CEM.

*Will be updated with CEM and MFC++ Agent 0.6.0

Links:

Source:

Advanced XML Processing with Apache NiFi 1.9.1

With the latest version of Apache NiFi, you can now directly convert XML to JSON or Apache AVRO, CSV or any other format supported by RecordWriters. This is a great advancement. To make it even easier, you don't even need to know the schema before hand. There is a built-in option to Infer Schema.

The results of an RSS (XML) feed converted to JSON and displayed in a slack channel.

Besides just RSS feeds, we can grab regular XML data including XML data that is wrapped in a Zip file (or even in a Zipfile in an email, SFTP server or Google Docs).

Get the Hourly Weather Observation for the United States

Decompress That Zip

Unpack That Zip into Files

One ZIP becomes many XML files of data.

An example XML record from a NOAA weather station.

Converted to JSON Automagically

Let's Read Those Records With A Query and Convert the results to JSON Records

Read RSS Feed of Travel Warnings

In this one simple example, we are ingesting all of the observed weather from all observation stations in the United States via one downloaded ZIP file containing all of the hourly XMLs. Apache NiFi can easily acquire this file, decompress it and unpack the files from the zip. We can then convert all of them as records by inferring their schema and building new records in the output format of our choice. We can then pull apart values we need or push this new cleaned format to one or more storage options including HBase, HDFS, Hive, SQL Database, MongoDB or elsewhere. We can also send this data on via Apache Kafka to a streaming engine such as Kafka Streams or Spark Structured Streaming for more processing or joining with other datasets.

QueryRecord is allowing us to write a SQL query such as SELECT * FROM FLOWFILE reading XML records and producing JSON records as a result. We can change fields or add things like SUMs and AVGs.

References:

Hourly Update

https://w1.weather.gov/xml/current_obs/all_xml.zip

A Look Around Apache NiFi 1.9.x

Apache NiFi 1.9 Deep Look

YARN From NiFi

YARN Job Kickoff

There are ways to integrate Apache NiFi with Apache YARN via YARN's APIs and command line tools.

IoT Series: Sensors: Utilizing Breakout Garden Hat: Part 1 - Introduction

IoT Series: Sensors: Utilizing Breakout Garden Hat: Part 1 - Introduction

An easy option for adding, removing and prototype sensor reads from a standard Raspberry Pi with no special wiring.

Hardware Component List:

Raspberry Pi
USB Power Cable
Pimoroni Breakout Garden Hat
1.12" Mono OLED Breakout 128x128 White/Black Screen
BME680 Air Quality, Temperature, Pressure, Humidity Sensor
LWM303D 6D0F Motion Sensor (X, Y, Z Axes)
BH1745 Luminance and Color Sensor
LTR-559 Light and Proximity Sensor 0.01 lux to 64,000 lux
VL53L1X Time of Flight (TOF) Sensor Pew Pew Lasers!

Software Component List:

Raspian
Python 2.7
JDK 8 Java
Apache NiFi
MiniFi

Source Code:

https://github.com/tspannhw/minifi-breakoutgarden

Shell Script (https://github.com/tspannhw/minifi-breakoutgarden/blob/master/runbrk.sh)
Python (https://github.com/tspannhw/minifi-breakoutgarden/blob/master/brk.py)

Summary

Our Raspberry Pi has a Breakout Garden Hat with 5 sensors and one small display. The display is showing the last reading and is constantly updating. For debugging purposes, it shows the IP Address so I can connect as needed.

We currently run via nohup, but when we go into constant use I will switch to a Linux Service to run on startup.

The Python script initializes the connections to all of the sensors and then goes into an infinite loop of reading those values and building a JSON packet that we send via TCP/IP over port 5005 to a listener. MiniFi 0.5.0 Java Agent is using ListenTCP on that port to capture these messages and filter them based on alarm values. If outside of the checked parameters we send them via S2S/HTTP(s) to an Apache NiFi server.

We also have a USB WebCam (Sony Playstation 3 EYE) that is capturing images and we read those with MiniFi and send them to NiFi as well.

The first thing we need to do is pretty easy. We need to plug in our Pimoroni Breakout Garden Hat and our 6 plugs.

You have to do the standard installation of Python, Java 8, MiniFi and I recommend OpenCV. Make sure you have everything plugged in securely and the correct direction before you power on the Raspberry Pi.

Download MiniFi Java Here: https://nifi.apache.org/minifi/download.html

Install Python PIP curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py

Install Breakout Garden Library wget https://github.com/pimoroni/breakout-garden/archive/master.zip

unzip master.zip

cd breakout-garden-master

sudo ./install.sh

Running In NiFi

First we build our MiniFi Flow:

We have two objectives: listen for TCP/IP JSON messages from our running Python sensor collector and gather images captured by the PS3 Eye USB Webcam.

We then add content type and schema information to the attributes. We also extract a few values from the JSON stream to use for alerting.

We extract: $.cputemp, $.VL53L1X_distance_in_mm, $.bme680_humidity, $.bme680_tempf

These attributes are now attached to our flowfile which is unchanged. We can now Route on them.

So we route on a few alarm conditions:

${cputemp:gt(100)}

${humidity:gt(60)}

${tempf:gt(80)}

${distance:gt(0)}

We can easily add more conditions or different set values. We can also populate these set values from an HTTP / file lookup.

If these values are met we send to our local Apache NiFi router. This can then do further analysis with the fuller NiFi processor set including TensorFlow, MXNet, Record processing and lookups.

Local NiFi Routing

For now we are just splitting up the images and JSON and sending to two different remote ports on our cloud NiFi cluster.

These then arrive in the cloud.

As you can see a list of the flow files waiting to be processed (I haven't written that part yet). As you can see we are getting a few a second, we could get 100,000 a second if we needed. Just add nodes. Instant scaling. Cloudbreak can do that for you.

In part 2, we will start processing these data streams and images. We will also add Apache MXNet and TensorFlow at various points on the edge, router and cloud using Python and built-in Deep Learning NiFi processors I have authored. We will also break apart these records and send each sensor to it's own Kafka topic to be processed with Kafka Streams, Druid, Hive and HBase.

As part of our loop we write to our little screen current values:

Example Record


{
  "systemtime" : "12/19/2018 22:15:56",
  "BH1745_green" : "4.0",
  "ltr559_prox" : "0000",
  "end" : "1545275756.7",
  "uuid" : "20181220031556_e54721d6-6110-40a6-aa5c-72dbd8a8dcb2",
  "lsm303d_accelerometer" : "+00.06g : -01.01g : +00.04g",
  "imgnamep" : "images/bog_image_p_20181220031556_e54721d6-6110-40a6-aa5c-72dbd8a8dcb2.jpg",
  "cputemp" : 51.0,
  "BH1745_blue" : "9.0",
  "te" : "47.3427119255",
  "bme680_tempc" : "28.19",
  "imgname" : "images/bog_image_20181220031556_e54721d6-6110-40a6-aa5c-72dbd8a8dcb2.jpg",
  "bme680_tempf" : "82.74",
  "ltr559_lux" : "006.87",
  "memory" : 34.9,
  "VL53L1X_distance_in_mm" : 134,
  "bme680_humidity" : "23.938",
  "host" : "vid5",
  "diskusage" : "8732.7",
  "ipaddress" : "192.168.1.167",
  "bme680_pressure" : "1017.31",
  "BH1745_clear" : "10.0",
  "BH1745_red" : "0.0",
  "lsm303d_magnetometer" : "+00.04 : +00.34 : -00.10",
  "starttime" : "12/19/2018 22:15:09"
}

NiFi Templates

Let's Build Those Topics Now


/usr/hdp/current/kafka-broker/bin/kafka-topics.sh --create --zookeeper localhost:2181 --replication-factor 1 --partitions 1 --topic bme680
/usr/hdp/current/kafka-broker/bin/kafka-topics.sh --create --zookeeper localhost:2181 --replication-factor 1 --partitions 1 --topic  bh17455
/usr/hdp/current/kafka-broker/bin/kafka-topics.sh --create --zookeeper localhost:2181 --replication-factor 1 --partitions 1 --topic  lsm303d
/usr/hdp/current/kafka-broker/bin/kafka-topics.sh --create --zookeeper localhost:2181 --replication-factor 1 --partitions 1 --topic  vl53l1x
/usr/hdp/current/kafka-broker/bin/kafka-topics.sh --create --zookeeper localhost:2181 --replication-factor 1 --partitions 1 --topic ltr559

Hopefully in your environment, you will be able to have 3, 5 or 7 replication factor and many partitions. I have one Kafka Broker so this is what we are starting with.

Reference