Oct 5, 2017

Moving data from Lake into Warehouse: the power of Streams


Edited: Oct 5, 2017


One of our clients has a large archive of data, in wildly varying schemas, dumped in a folder structure in Azure Data Lake. They're split up by date and time, but every folder contains gzipped CSVs with wildly varying structures and sizes. To make that data useful, we needed to ingest that data into an Azure Data Warehouse.


The default way to do this in Data Warehouse would be to use PolyBase. It can handle Gzipped CSVs quite well out of the box, and it's very efficient for the larger files. In our case, some of the files are several gigabytes, so using polybase is a natural fit for those.

However, there are also thousands of smaller files, from a few kilobytes up to hundreds of megabytes. For those, especially the small ones, polybase requires too much setup for the payoff. So we went looking for a better way, preferably one that is also reusable for Azure SQL DB (which doesn't support polybase).


One thing you can do in both DB and DW is bulk copying. Specifically, the .Net Bulk Copy Client can read data from any DataReader and stream it in to a table without keeping the entire file in memory, in a safe and fast way. A DataReader feeds the Bulk Copy Client line by line, value by value, all of the data. Unfortunately, there are no DataReaders that read directly from Data Lake, let alone Gzipped CSV files, so we decided to write that capability ourselves.


We know that lots of usable packages for data manipulation can handle Streams. If we could read the Data Lake files with Streams, we could feed them through some steps and then encapsulate the last step as a DataReader ourselves. This would give us the ability to go from Data Lake into DW without needing to do any file copying or keeping an entire file in memory.


We initially tried working with the Stream the Data Lake client itself can give us, but we found that it was forcibly disconnected often. It just isn't designed for streaming, but rather for downloading onto disk or memory and then disposing of the stream. However, Data Lake also provides the ability to read chunks of the data as bytes, using just offset and count. That closely mirrors what a Stream can do. We ended up writing our own Stream implementation with position tracking, that reads chunks from Data Lake with the correct offset whenever it is read from. We even added some read-ahead logic to make better use of the inbound bandwidth. This class also handles reconnection and retries for DL.


Now that we have a Stream, we could start wiring this with some other packages. .Net's BufferedStream over the top allows us to decide how large the chunks we read should be, with a lot of automated management. A GzipStream (from System.IO.Compression) will read that BufferedStream and decompress it to its proper bytes for us. After that, we used a StreamReader that we feed into a CsvReader (from the CsvHelper package, a great little package for this kind of stuff). We do some set up to read the first row to determine columns, but aside from that the CsvReader gives us all the power we need to implement an IDataReader with it. Now we have a full chain of streams, from reading bytes in Data Lake, to buffering, decompressing, reading as CSV and pumping the value into a table in Data Warehouse. All without keeping anything larger than a few MB buffer in memory!


It's important to note that this method is fast, but will never beat PolyBase for raw speed. Bulk Copy is always restricted by the speed of the DW Control Node, which PolyBase handily sidesteps by distributing the operations across all nodes. Additionally, due to Bulk Copy transactions taking up to 15 seconds to commit at the end of streaming, and other speed limits in how quickly we can decompress a file, it's important to run copies for as many files at a time as possible. The more you can hit the DW with at the same time, the more throughput you can get.


To make the application a bit more powerful, we added automatic creation of destination tables, some logic to handle files with missing columns, and an automatic fallback to polybase for the largest of files.


All in all, it's been an interesting bit of development, and has definitely shown the power of writing and composing Streams. Because they're so widely supported by other packages, mainly with the intention of using them to read large files from disk, we can use them to set up intricate but very efficient flows of data. Give it a try!

New Posts
  • Something that comes up quite frequently when people start using Spark is "How can I filter my DataFrame using the contents of another DataFrame?". People with SQL experience will immediately look to trying to replicate the following. SELECT * FROM table_a a WHERE EXISTS (SELECT * FROM table_b b WHERE b.Id = a.Id) So how do you do this in Spark? Well, some people will try to use the Column.isin method which uses varargs, this is okay for a small set of values but if you have a couple of large DataFrames then it's less than optimal as each row needs to be evaluated against the list. So what's the other choice? We can use joins to do the same thing. There are 2 we can use, a SEMI JOIN which is equivalent to our above example of running EXISTS; the other is ANTI JOIN which is equivalent to a NOT EXISTS. Using the above example and keeping the table names as DataFrame names we could re-write this in Scala as: table_a.join(table_b, Seq("Id"), "left_semi") These 2 joins are unique in that they only return the output of the left DataFrame, without any content from the right DataFrame. So what does this look like in practice. Well using Azure Databricks we can quickly create some sample data to try them out. First lets create a couple of DataFrames. First lets runs a simple query to find heroes which have an arch-enemy. This uses the SEMI JOIN to keep records in the left DataFrame where there is a matching record in the right DataFrame. Now, lets have look for heroes who've been a little more active and have removed their arch-enemies (for now). This time we've used an ANTI JOIN to keep only those records in the left DataFrame where there are no matching records in the right DataFrame. You'll notice that in the examples the join condition uses the slightly longer form, that's because in this example the columns we're joining on have different names, and also because there is a column in both DataFrames which have the same name.
  • Recently I needed to deploy an Azure Data Lake Store - Gen 2 instance and thought I'd take the opportunity to use some custom ARM template functions . These aren't something you often see in the example templates but can be really useful if there's a complex expression which you find yourself writing repeatedly within a template. If, for instance, you routinely create resource names based on a prefix, unique name and a suffix then this could save you a few keystrokes. In essence you are simply parameterizing the expression as follows: In this way you can use this simpler expression where you would have previously used the more complex version. [namespace.function(parameter1, parameter2)] If you want to see what this looks like in a full template then checkout this simple ARM template I put together for creating a Data Lake Store - Gen 2 instance over on GitHub.
  • Documentation is not something people often spend time reading, or if they do then its to quickly find the one thing their after and then get out as quickly as possible, very similar to how I do my Christmas shopping. Sometimes it's worth spending time reading the documentation though as there can be some useful bits of information hidden in summary descriptions, links etc... One such item is the Azure Data Lake Store client. If you find yourself reading or writing a lot of files and your doing it in multiple tasks (or threads, but you should be using Tasks if possible), then reading the docs can really help you out. For instance this snippet taken from the description at the top of the documentation page . If an application wants to perform multi-threaded operations using this SDK it is highly recomended to set ServicePointManager.DefaultConnectionLimit to the number of threads application wants the sdk to use before creating any instance of AdlsClient. By default ServicePointManager.DefaultConnectionLimit is set to 2. Okay, so how bad can things be if you don't read this? Well, to answer that I created an ADLS instance and uploaded a number of small parquet files. Then wrote an application to read each file (using the excellent Parquet .NET ) and return the number of records in the file, each file is processed in it's own Task and each uses the same AdlsClient instance. The simple process being followed here is to get a list of files, call " ProcessPath " on each and then when all the files have been process output the results. The output of this initial version is as follows: It's not too bad, but with multiple tasks I would have expected it to be better. Looking at the documentation snippet above it suggests we need to change the ServicePointManager.DefaultConnectionLimit value, but what to? Well doing some digging around came across a suggestion from Microsoft Support which, for ASP.NET, is to limit the number of requests that can execute at the same time to 12 per CPU (or 12 per core). So let's give that a go and see what happens. The code change for this is pretty simple and we can use System.Environment to get the number of processors available. So does it make much of a difference? Well, yes, quite a lot of difference actually. I ran the code in both variations a few more times to check it wasn't intermittent networking issues, other processes on my laptop interfering etc... but no, it really does make that much of a difference. So next time you're working with multiple tasks sharing resources, maybe spend a bit of time reading the documentation to see if there's anything which can make a difference to your application.