Ust

Ust Oldfield's Blog

Archiving the Data Lake

In a blog introducing the Data Lake Framework, keen readers will be aware that in the diagram there’s a box titled “ARCHIVE” but it has not been brought up since. The reason why the Archive layer in the data lake has not been discussed is because we’ve been waiting for the Archive Tier in Blob Storage.

To remind readers of the framework and where the archive layer sits in it, here it is again with the archive layer highlighted.

image

The Archive Blob

The Archive access tier in blob storage was made generally available today (13th December 2017) and with it comes the final piece in the puzzle to archiving data from the data lake.

Where Hot and Cool access tiers can be applied at a storage account level, the Archive access tier can only be applied to a blob storage container. To understand why the Archive access tier can only be applied to a container, you need to understand the features of the Archive access tier. It is intended for data that has no or low SLAs for availability within an organisation and the data is stored offline (Hot and Cool access tiers are online). Therefore, it can take up to 15 hours for data to be made online and available. Brining Archive data online is a process called rehydration (fitting for the data lake). If you have lots of blob containers in a storage account, you can archive them and rehydrate them as required, rather than having to rehydrate the entire storage account.

Archive Pattern

An intended use for the Archive access tier is to store raw data that must be preserved, even after it has been fully processed, and does not need to be accessed within 180 days.

Data gets loaded into the RAW area of the data lake, is fully processed through to CURATED, and a copy of the raw data is archived off to a blob container with a Cool access tier applied to it. When the archive cycle comes about, a new Cool access tiered blob container is created and the now old container has its access tier changed to Archive.

For example, our Archive cycle is monthly and we have a Cool access tiered blob container in our storage account called “December 2017”. When data has finished being processed in the Azure Data Lake, the Raw data is archived to this blob container. January comes around, we create a new blob container called “January 2018” with Cool access tier settings and change the access tier of “December 2017” from Cool to Archive.

This data has now been formally achieved and is only available for disaster recovery, auditing or compliance purposes. 




LETS Process Data–Modern ETL for the Data Lake

At the PASS Summit this year, I attended a session by Michael Rys. In this session he introduced the concept of LETS as an approach to process data in the data lake. If you are familiar with data lake, then you will be familiar of having to apply a schema to the data held within. The LETS approach is purpose design for schematization.

Where ETL stands for Extract, Transform, Load or ELT stands for Extract, Load, Transform – LETS stands for Load, Extract, Transform, Store.

Data are Loaded into the data lake

Data are Extracted and schematized

Data are Transformed in rowsets

Data are Stored in a location, such as the Catalog in Azure Data Lake Analytics, Azure Data Warehouse, Azure Analysis Services, for analysis purposes.

image

I really like this approach as it makes sense for how data are handled in the data lake. It’s something that I will be advocating and using, and I hope you do too!



Data Flow: Job Execution in the Azure Data Lake

In this blog we will deep dive into the job execution of the Azure Data Lake Analytics (ADLA). If you’ve used ADLA you will have come across the following image:

Picture1

This is the graphical representation of the job execution. But what is is doing at each stage?

Preparing

The job execution begins with the authoring of the U-SQL script. The U-SQL script itself is the logical plan of how you intend to transform input data into output data. The script gets compiled, which will translate your U-SQL script primarily into C#, as well as a few other items such as XML– which will contain metadata information and the job graph. Once compiled it will create an initial plan – which will then be optimised thus producing an optimised plan. If it fails at this stage it will mostly be down to your U-SQL script failing to compile successfully. Fortunately, the error codes at this stage are fairly helpful and will indicate the line in question.

Queued

Once compiled and optimised the job progresses to the queued stage. There are a few things which can cause the a job to queue and these are:

  • Running Jobs
    • The default setting per subscription is 3. So if you or someone else in your subscription is already running a series of jobs, your job will be queued until at least one of them completes
  • Higher priority jobs
    • As the queue is ordered by job priority, with loser numbers having a higher priority, other jobs can jump to the top of the queue thus forcing your job to queue for longer
  • Lack of resources
    • Even if there are no running jobs and there are no jobs ahead of yours in the queue, your job will continue to queue if there are not enough Azure Data Lake Analytic Units (ADLAUs) to start the job.

 

Running

Once the job has finished queuing it will run. The Job Manager allocates vertexes, which are collections of data to be processed, to the ADLAUs and uses YARN to orchestrate it. The vertexes will then execute. At this stage, if there is a vertex failure you can download the vertex locally and debug in visual studio.

When the vertexes have successfully completed you will be able to consume the end product – either locally, if you have run the job locally, or in the Azure Data Lake Store.

Conclusion

Job Execution

The above process can be visualised in this graph. Knowing how ADLA executes your job is the first of many steps to being able to write performant U-SQL and debug effectively. Be on the look out for more blogs on the technical deep dive into ADLA!

Testing the Waters: An Overview of Data Science using Azure Data Lakes

Data Science can fit seamlessly within the ecosystem of the data lake, whether this is through HDInsight or the extensibility of Azure Data Lake Analytics and U-SQL. This blog will give a brief overview of what Data Science is; how to link Data Science toolkits to the Azure Data Lake; and best practices for managing the data output from experiments.

 

Data Science

Data Science is the relatively new kid on the block. One way to consider data science is as an evolutionary step in interdisciplinary fields like business analysis that incorporate computer science, modelling, statistics, analytics, and mathematics.
At its core, data science involves using automated methods to analyse massive amounts of data and to extract insight from them. Data science is helping to create new branches of science, and influencing areas of social science and the humanities. The trend is expected to accelerate in the coming years as data from mobile sensors, sophisticated instruments, the web, and more, grows.

 

Data Science In The Data Lake

The nature of the Azure Data Lake Store lends itself to Data Science in that it can hold any data, which the data scientist will want to access, transform and analyse.

 

HDInsight contains many implementations for data science, such as Spark, R Server and others. Hooking HDInsight to Azure Data Lake Store is pretty simple and follows these steps:

In the Azure Portal Marketplace, select HDInsight which will bring up a series of blades. In this blog, I will be using Spark as my cluster type on HDInsight.

image

In the storage settings you can then link your HDInsight cluster to Azure Data Lake Store

image

Confirm your configuration on the next blade and wait around 20 minutes for your cluster to deploy and you’re good to go!

 

With Azure Data Lake Analytics, you incorporate data science by extending the capabilities of U-SQL and you do this by installing a series of files.

Open up your Azure Data Lake Analytics account and click on Sample Scripts at the top. This will bring forward the following blade

image

From there you’ll want to click on the U-SQL Advanced Analytics tab, which will copy about 1.5GB of files to the default Azure Data Lake Store associated to your ADLA account. This will take about 3 minutes to complete.

When it’s finished copying the files it will then call a job to register the extension, which can be found in the assemblies folder of the master database.

image

More resources about extensibility of U-SQL can be found here:

https://blogs.msdn.microsoft.com/azuredatalake/2017/03/10/using-custom-python-libraries-with-u-sql/

https://blogs.msdn.microsoft.com/azuredatalake/2016/11/22/u-sql-advanced-analytics-introducing-python-extensions-for-u-sql/

The Laboratory

Within Azure Data Lake Store, Folder and File Management is incredibly important for a well running data lake. See my blogs on Storage and Best Practices and Shaping The Lake for more information on how to set up your Azure Data Lake Store.

The Laboratory is an area to be exclusively used by a data scientist. It’s an area where they can persist the results of experiments and data sets without impacting the day-to-day operations within the data lake or other data scientists. The laboratory is organised in to two broad area: Desks and Exhibits.

Desks contain personal workspaces, the contents of which can be organised however the person wishes. It can be as well organised, or disorganised, as the person themselves. The Exhibit contains data sources produced in the Laboratory which are ready to be consumed by other users or systems. Both of which are laid out below.

image

As always, if you have any feedback or comments do let me know!

Azure Data Lake Store–Storage and Best Practices

The Azure Data Lake Store is an integral component for creating a data lake in Azure as it is where data is physically stored in many implementations of a data lake. Under the hood, the Azure Data Lake Store is the Web implementation of the Hadoop Distributed File System (HDFS). Meaning that files are split up and distributed across an array of cheap storage.

 

What this blog will go into is the physical storage of files in the Azure Data Lake Store and then best practices, which will utilise the framework.

 

Azure Data Lake Store File Storage

As mentioned, the Azure Data Lake Store is the Web implementation of HDFS. Each file you place into the store is split into 250MB chunks called extents. This enables parallel read and write. For availability and reliability, extents are replicated into three copies. As files are split into extents, bigger files have more opportunities for parallelism than smaller files. If you have a file smaller than 250MB it is going to be allocated to one extent and one vertex (which is the work load presented to the Azure Data Lake Analytics), whereas a larger file will be split up across many extents and can be accessed by many vertexes.

 

The format of the file has a huge implication for the storage and parallelisation. Splittable formats – files which are row oriented, such as CSV – are parallelizable as data does not span extents. Non-splittable formats, however, – files what are not row oriented and data is often delivered in blocks, such as XML or JSON – cannot be parallelized as data spans extents and can only be processed by a single vertex.

 

In addition to the storage of unstructured data, Azure Data Lake Store also stores structured data in the form of row-oriented, distributed clustered index storage, which can also be partitioned. The data itself is held within the “Catalog” folder of the data lake store, but the metadata is contained in the data lake analytics. For many, working with the structured data in the data lake is very similar to working with SQL databases.

 

Azure Data Lake Store Best Practices

The best practices generally involve the framework as outlined in the following blog: http://blogs.adatis.co.uk/ustoldfield/post/Shaping-The-Lake-Data-Lake-Framework

The framework allows you to manage and maintain your data lake. So, when setting up your Azure Data Lake Store you will want to initially create the following folders in your Root

image

Raw is where data is landed in directly from source and the underlying structure will be organised ultimately by Source.

image

Source is categorised by Source Type, which reflects the ultimate source of data and the level of trust one should associate with the data.

Within the Source Type, data is further organised by Source System.

image

Within the Source System, the folders are organised by Entity and, if possible, further partitioned using the standard Azure Data Factory Partitioning Pattern of Year > Month > Day etc., as this will allow you to achieve partition elimination using file sets.

image

The folder structure of Enriched and Curated is organised by Destination Data Model. Within each Destination Data Model folder is structured by Destination Entity. Enriched or Curated can either be in the folder structure and / or within the Database.  

image

Shaping The Lake: Data Lake Framework

The Azure Data Lake has just gone into general availability and the management of Azure Data Lake Store, in particular, can seem daunting especially when dealing with big data. In this blog, I will take you through the risks and challenges of working with data lakes and big data. Then I will take you through a framework we’ve created to help best manage these risks and challenges.

If you need a refresh as to what a data lake is and how to create your first Azure Data Lake Store and your first Azure Data Lake Analytics job, please feel free to follow the links.

Risks and Challenges of Big Data and Data Lake

The challenges posed by big data are as follow:

  • Volume – is the sheer amount of data becoming unmanageable?
  • Variety – Structured tables? Semi-structured JSON? Completely unstructured text dumps? We can normally manage with systems that contain just one of these, but if we’re dealing with a huge mix, it gets very tricky
  • Velocity – How fast is the data coming in? And how fast do we need to get it to the people who need it?
  • Veracity - How do we maintain accuracy, veracity, when the data is of varying volumes, the sources and structures are different and the speed in which they arrive in the Lake are of differing velocities?

Managing all four simultaneously is where the challenges begin.

It is very easy to treat a data lake as a dumping ground for anything and everything. Microsoft’s sale pitch says exactly this – “Storage is cheap, Store everything!!”. We tend to agree – but if the data is completely malformed, inaccurate, out of date or completely unintelligible, then it’s no use at all and will confuse anyone trying to make sense of the data. This will essentially create a data swamp, which no one will want to go into. Bad data & poorly managed files erode trust in the lake as a source of information. Dumping is bad.

There is also data drowning – as the volume of the data tends towards the massive and the velocity only increases over time we are going to see more and more information available via the lake. When it gets to that point, if the lake is not well managed, then users are going to struggle to find what they’re after. The data may all be entirely relevant and accurate, but if users cannot find what they need then there is no value in the lake itself. Essentially, data drowning is when the amount of data is so vast you lose the ability to find what’s in there.

If you ignore these challenges, treat the lake like a dumping ground, you will have contaminated your lake and it will no longer be fit for purpose.

If no one uses the Data Lake, it’s a pointless endeavour and not worth maintaining.

Everyone needs to be working together to ensure the lake stays clean, managed and good for a data dive!

Those are the risks and challenges we face with Azure Data Lake. But how do we manage it?

The Framework

Data Lake

 

We’ve carved the lake up into different sections. The key point is that the lake contains all sorts of different data – some that’s sanitised and ready to consume by the business user, some that’s indecipherable raw data that needs careful analysis before it is of use. By ensuring data are carefully managed you can instantly understand the level of preparation that data has undergone.

Data flows from left to right – the further left areas represent where data has been input directly from source systems. The horizontal sections describe the level of preparation – Manual, Stream and Batch.

Manual – aka, the laboratory. Here data is manually prepared with ad-hoc scripts.

Stream – The data here flows in semi-real time, coming from event hubs and being landed after processing through stream-specific tools such as Streaming Analytics. Once landed, there is no further data processing – the lake is essentially a batch processing tool.

Batch – This is more traditional data processing, the kind of “ETL” seen by many BI developers. We have a landing area for our raw data, a transitional area where data is cleaned, validated, enriched and augmented with additional sources and calculation, before finally being placed in a curated area where it is ready for consumption by the business.

We’re taking the blank-canvas of the Data Lake Store and applying a folder structure, a file management process and a curation process over the top.

The folder structure itself can be as detailed as you like, we follow a specific structure ourselves:

Data Lake 2

 

The Raw data area, the landing place for any files entering the lake, has sub-folders for each source of data. This allows for the easy browsing of the data sources within the Lake and ensures we are not receiving the same data twice, even if we use it within different systems.

The Enriched and Curated layers however, have a specific purpose in mind. We don’t take data and enrich/clean/process it without a business driver, it’s not something we do for fun. We can therefore assign a project or system name to it, at this point it is organised into these end-systems. This means we can view the same structure within Enriched as within Curated.

Essentially Raw data is categorised by Source whilst Enriched and Curated data is categorised by Destination.

There’s nothing complicated about the Framework we’ve created or the processes we’ve ascribed to it, but it’s incredibly important that everyone is educated on the intent of it and the general purpose of the data lake. If one user doesn’t follow process when adding data, or an ETL developer doesn’t clean up test files, the system starts to fall apart and we succumb to the challenges we discussed at the start.

To summarise, structure in your Azure Data Lake Store is key to maintaining order:

• You need to enforce and maintain folder structure.

• Remember that structure is necessary whether using unstructured data or tables & SQL

• Bear in mind that schema on read applies temporary structure – but if you don’t know what you’re looking at, this is going to be very hard to do!