Handling Rogue Queries In Spark

Alicja Luszczak, et al, introduce the Query Watchdog:

The previous query would cause problems on many different systems, regardless of whether you’re using Databricks or another data warehousing tool. Luckily, as an user of Databricks, this customer has a feature available that can help solve this problem called the Query Watchdog.

Note: Query Watchdog is available on clusters created with version 2.1-db3 and greater.

A Query Watchdog is a simple process that checks whether or not a given query is creating too many output rows for the number of input rows at a task level. We can set a property to control this and in this example we will use a ratio of 1000 (which is the default).

It looks like this is an all-or-nothing process, but a very interesting start.

Real-Time Weather With HDF

Balaji Kandregula shows how to use Hortonworks Data Flow components to process weather events in real time:

It’s live weather reporting using HDF, Kafka, and Solr.

Here are the environment requirements for implementing:

  • HDF (for HDF 2.0, you need Java 1.8).
  • Kafka.
  • Spark.
  • Solr.
  • Banana.

Now let’s get on to the steps!

There are a lot of moving parts there, but the pieces do plug in well enough and there are a lot of screen shots to guide you along the way.

PySpark Persistence

David Crook shows how to save data to disk from PySpark:

This is working on HDInsight v3.5 w/Spark 2.0 and Azure Data Lake Storage as the underlying storage system.  What is nice about this is that my cluster only has access to its cluster section of the folder structure.  I have the structure root/clusters/dasciencecluster.  This particular cluster starts at dasciencecluster, while other clusters may start somewhere else.  Therefor my data is saved to root/clusters/dasciencecluster/data/open_data/RF_Model.txt

It’s pretty easy to do, and the Scala code would look suspiciously similar.  The Java version of the code would be seven pages long.

Spark Deep Learning On AWS

Joseph Spisak, et al, show how to configure and use BigDL in Amazon Web Services’s ElasticMapReduce:

Classify text using BigDL

In this tutorial, we demonstrate how to solve a text classification problem based on the example found here. This example uses a convolutional neural network to classify posts in the 20 Newsgroup dataset into 20 categories.

We’ve provided a companion Jupyter notebook example on GitHub that you can open in the Jupyter dashboard to execute the code sections.

There’s a lot to this tutorial.

Text Normalization With Spark

Engineers at Treselle Systems have put together a two-part series on text normalization using Apache Spark.  First, they walk through normalizing the text:

We have used Spark shared variable “broadcast” to achieve distributed caching. Broadcast variables are useful when large datasets need to be cached in executors. “stopwords_en.txt” is not a large dataset but we have used in our use case to make use of that feature.

What are Broadcast Variables?
Broadcast variables in Apache Spark is a mechanism for sharing variables across executors that are meant to be read-only. Without broadcast variables, these variables would be shipped to each executor for every transformation and action, which can cause network overhead. However, with broadcast variables, they are shipped once to all executors and are cached for future reference.

From there, they dig into details on what the Spark engine did and why we see what we do:

Note: Stage 2 has both reduceByKey() and sortByKey() operations and as indicated in job summary “saveAsTextFile()” action triggered Job 2. Do you have any guess whether Stage 2 will be further divided into other stages in Job 2? The answer is: yes Job 2 DAG: This job is triggered due to saveAsTextFile() action operation. The job DAG clearly indicates the list of operations used before the saveAsTextFile() operations.Stage 2 in Job 1 is further divided into another stage as Stage 2. In Stage 2 has both reduceByKey() and sortByKey() operations and both operations can shuffle the data so that Stage 2 in Job 1 is broken down into Stage 4 and Stage 5 in Job 2. There are three stages in this job. But, Stage 3 is skipped. The answer for the skipped stage is provided below “What does “Skipped Stages” mean in Spark?” section.

There’s some good information here if you want to become more familiar with how Spark works.

The Basics Of SparkR

Yanbo Liang has an introductory article on what SparkR is and why you might want to use it:

However, data analysis using R is limited by the amount of memory available on a single machine and further as R is single threaded it is often impractical to use R on large datasets. To address R’s scalability issue, the Spark community developed SparkR package which is based on a distributed data frame that enables structured data processing with a syntax familiar to R users. Spark provides distributed processing engine, data source, off-memory data structures. R provides a dynamic environment, interactivity, packages, visualization. SparkR combines the advantages of both Spark and R.

In the following section, we will illustrate how to integrate SparkR with R to solve some typical data science problems from a traditional R users’ perspective.

This is a fairly introductory article, but gives an idea of what SparkR can accomplish.

Using h2o.ai On HDInsight

Xiaoyong Zhu shows how to set up h2o.ai on Azure HDInsight:

H2O Flow is an interactive web-based computational user interface where you can combine code execution, text, mathematics, plots and rich media into a single document, much like Jupyter Notebooks. With H2O Flow, you can capture, rerun, annotate, present, and share your workflow. H2O Flow allows you to use H2O interactively to import files, build models, and iteratively improve them. Based on your models, you can make predictions and add rich text to create vignettes of your work – all within Flow’s browser-based environment. In this blog, we will only focus on its visualization part.

H2O FLOW web service lives in the Spark driver and is routed through the HDInsight gateway, so it can only be accessed when the spark application/Notebook is running

You can click the available link in the Jupyter Notebook, or you can directly access this URL:

https://yourclustername-h2o.apps.azurehdinsight.net/flow/index.html

Setup is pretty easy.

Probabilistic Record Linking In Spark

Tom Lous builds a solution to link similar companies together by address:

Recently a colleague asked me to help her with a data problem, that seemed very straightforward at a glance.
She had purchased a small set of data from the chamber of commerce (Kamer van Koophandel: KvK) that contained roughly 50k small sized companies (5–20FTE), which can be hard to find online.
She noticed that many of those companies share the same address, which makes sense, because a lot of those companies tend to cluster in business complexes.

Read on for the solution.  Like many data problems, it turns out to be a lot more complicated than you’d think at first glance.

Rolling A Log Analytics System

Michael Sun and Jeff Shmain put together a log analytics sytem using several technologies:

This is an example of tiered system design. Tiered system is a system design pattern where data is categorized and stored in different data stores that match best to each category. It can both improve performance and lower the cost of a system. One of the most famous tiered system designs is computer memory hierarchy.  In the log analytics use case, analysts mostly search for logs in recent months, but often run batch jobs to get long term trends from logs in recent years. Therefore, recent logs are indexed and stored in Solr for search, while years of logs are stored in HBase for batch processing. As such, the index in Solr is small, which both improves performance and reduces cost, among other benefits.

Although only months of logs are stored in Solr, the logs before that period are stored in HBase and can be indexed on demand for further analysis.

Now that we have covered a high level architecture of a log analytics system, we will dive into more details of individual components.

This looks like a solid architecture for a logging system and can apply to other cases as well.

Tuning Kafka And Spark Data Pipelines

Larry Murdock explains the tuning options available to Kafka and Spark Streams:

Kafka is not the Ferrari of messaging middleware, rather it is the salt flats rocket car. It is fast, but don’t expect to find an AUX jack for your iPhone. Everything is stripped down for speed.

Compared to other messaging middleware, the core is simpler and handles fewer features. It is a transaction log and its job is to take the message you sent asynchronously and write it to disk as soon as possible, returning an acknowledgement once it is committed via an optional callback. You can force a degree of synchronicity by chaining a get to the send call, but that is kind of cheating Kafka’s intention. It does not send it on to a receiver. It only does pub-sub. It does not handle back pressure for you.

I like this as a high-level overview of the different options available.  Definitely gets a More Research Is Required tag, but this post helps you figure out where to go next.

Categories

April 2017
MTWTFSS
« Mar  
 12
3456789
10111213141516
17181920212223
24252627282930