.NET for Apache Spark v2.1.2 in .NET7 ?
GeorgeS2019 opened this issue · comments
Update 28thFeb2023
Thanks to @dragorosson
Following the advice by @dragorosson
- Building Spark .NET
- Building Spark .NET Scala Extensions Layer
It is possible in Windows to prepare openjdk-8-jdk, mvm, spark-3.2.3-bin-hadoop3.2 to create
version = 2.1.1
microsoft-spark-3-2/target/microsoft-spark-3-2-3_2.12-[2.1.1].jar
By compiling Microsoft.Spark.Worker in .NET7, instead of .NET6.0, this ensure consistency of addressing the BinaryFormatter compiling error. UDF now works. Delta works.
Next open issue:
- Key issue left is support for
microsoft-spark-3-3.jar
Previous discussions
#### Update 23rdFeb2023Please share your feedback to this observation provided here, but please do this within this issue for tracking purposes.
For Spark, this project is key for .NET developers to stay within .NET when dealing with big data analytics. It is unclear WHY there are questionable and sporadic commitments shown here. If this effort here fails OR with further delay, it could have a ripple effect on the ENTIRE machine learning and deep learning .NET efforts.
The triangular THREE PRONGED .NET efforts to KEEP big data analytics within .NET could be questionable.
- Machine Learning (ML.NET)
- PolyGlot
- .NET for Spark
Update 20thFeb2023
- Test the WIP .NET6 version
- One of the main TODO for the (yet-to-release) .NET6 is UDF support.
BinaryFormatter problem has been reported since Dec 2020. This will be the top priority for the .NET6 before it is being released
- One of the main TODO for the (yet-to-release) .NET6 is UDF support.
- Check the WIP .NET6 with
microsoft-spark-3-2.jar- Check the WIP .NET6 with
microsoft-spark-3-2_2.12-2.1.1.jarSo far, I could only get the WIP merged .NET6 to work with
microsoft-spark-3-1_2.12-2.1.1.jar
- Check the WIP .NET6 with
Update 15thFeb2023
It seems the Azure Synapse team has officially deleted ALL C# .NET for Spark samples for Synapse Jan 2023. Sad!
It seems the support for .NET for Spark is back :-) We will be getting the .NET6 version soon.
Anyway, what has changed your mind @GeorgeS2019? The fact that there're recent PRs and commits, how is this demonstrating that Microsoft still supports .NET for Spark? I'm really worried about this, because we adopted .NET for Spark few months ago.
The more feedback and participation from .NET for Spark users, the more commitment we could expect from the dotnet team.
Reading through the issues, PRs, and discussions here, it seems to me that the dotnet team needs MORE participation and feedback to justify MORE commitment to this project.
Nothing to be alarming!
We need more active Participation from users HERE! => put some emoji, here and there, raise some feedback WILL HELP to keep the dotnet team continue the support for .NET for SPARK
Questions for ALL!
How many of you are able to get UDF to work? if so, which version of Microsoft Spark e.g. microsoft-spark-3-2.jar, which version of Microsoft.Spark.dll
I am able to use UDFs. I'm compiling Microsoft.Spark.Worker with .Net 7 and would be using the jar from the NuGet package except I needed to run with Spark 3.2.2 not 3.2.1 so I had to build the jars myself. I use the latest NuGet package and copy the custom jar into the output directory whenever I build. The main issue I ran into was dependency resolution but compiling Microsoft.Spark.Worker and my spark program with dotnet publish --self-contained has solved that.
@dragorosson
I am still learning.
I needed to run with Spark 3.2.2 not 3.2.1 so I had to build the jars myself.
microsoft-spark-3-2_2.12-2.1.1.jar is build with the latest commit when running mvn clean package
microsoft-spark-3-2_2.12-2.1.1.jar is distributed with nuget.
Sorry for my ignorance:
- why the provided
microsoft-spark-3-2_2.12-2.1.1.jarfrom nuget does not allow you to run Spark 3.2.2
object DotnetRunner extends Logging {
private val DEBUG_PORT = 5567
private val supportedSparkMajorMinorVersionPrefix = "3.2"
private val supportedSparkVersions = Set[String]("3.2.0", "3.2.1", "3.2.2", "3.2.3")
-
does the newly build
microsoft-spark-3-2_2.12-2.1.1.jarsupport 3.2.3 ? -
What need to be done to create
microsoft-spark-3-3folder in thespark\src\scalafolder
What are the additional modifications needed to bring the existing spark-3.3 to be compatible with the
microsoft-spark-3-3?
Ref:
Support for 3.2.2
Support for 3.2.3
* why the provided `microsoft-spark-3-2_2.12-2.1.1.jar ` from nuget does not allow you to run Spark 3.2.2
The most recent nuget package is from 9 months ago https://www.nuget.org/packages/Microsoft.Spark/2.1.1
And the change allowing 3.2.2 was merged 1 month ago #1122
The .Net for Spark team will need to do a new release.
* does the newly build `microsoft-spark-3-2_2.12-2.1.1.jar ` support 3.2.3 ?
If you just built it locally, yes, it should. If you try to spark-submit with an unsupported version, it will fail quickly and tell you want versions it does support.
* What need to be done to create `microsoft-spark-3-3` folder in the` spark\src\scala` folder
I don't know. Spark 3.3 is not supported yet so I don't think it'd produce that jar.
What are the additional modifications needed to bring the existing spark-3.3 to be compatible with the
microsoft-spark-3-3
I have no idea. I don't have time to dig around the internals or look through previous PRs that added support for previous versions, etc.!
What changes have you made to overcome
#pragma warning disable SYSLIB0011 just like got merged in with the .Net 6 PR
As someone that has never used Apache Spark and seeing this big warning in the first page of the docs, and then seeing this issue, I do not feel confident in using it if it's always going to be so far behind it needs a special warning...
.NET for Apache Spark targets an out of support version of .NET (.NET Core 3.1). For more details see the .NET Support Policy.
@lloydjatkinson
FYI, The "Synapse Analytics" PG has started putting that warning all over the place. It even appears in some docs that have absolutely nothing to do with Synapse.
Technically it is not even true, since the .Net 6 PR is in place (see #1112 ) At this point it appears that the communication is willfully incorrect.
What happened is that I had run into a bug on the Synapse-Spark platform, which did NOT occur on any other Spark platform. The root cause of the bug was not actually even related to .Net, in the final analysis. Instead of working on the investigation and working on the fix for the bug, this Synapse PG decided they didn't want to support .Net anymore. So they killed .Net in Synapse 3.3, and they started down the scorched-earth path to make sure nobody ever tries to use .Net-for-Spark on any of the competing cloud platforms either (eg. Databricks or HDI).
Given that they are now making people fearful about this project, they are certainly accomplishing their mission!
I've used .Net for Spark many years and it is a beautiful thing. The PR for .Net 6 is going to allow us to look forward to many more years. This project basically allows .Net to piggyback on the Python bindings, and prevents the need for a .Net programmer to resort to another (inferior) language in order to do our MPP data transformations and ETL's.
OK this goes over by head a bit and sounds a bit political? Is Microsoft or the .NET team aware of this? What is the plan going forward?
Hi Team: Any timeline for the above PR and possible new release of this library. We are heavily invested in this library (because of our existing .Net dependency) and has been waiting to see a full compatibility of .Net 6.0 & Spark 3.2 (or above) from longtime! I have raised couple of feature requests (#958, #983, #1100) a while back but not much traction on those. Could you please let us know if this library is going to be supported going forward and a possible timeframe for an updated version.
cc @imback82
Hi @imback82 , @Niharikadutta , @GeorgeS2019 , @dbeavon,
Could you please let me know if we can depend on this library and if there's any chance we see a new version of this library at all?
Thanks
Hi @imback82 , @Niharikadutta , @GeorgeS2019 , @dbeavon, Could you please let me know if we can depend on this library and if there's any chance we see a new version of this library at all? Thanks
^ + @suhsteve @AFFogarty
@dragorosson
I hope everyone here is clear. I am independent, with nothing to do with Dotnet or Microsoft.
I am no longer maintaining this repo (no write access), but I think @AFFogarty / @suhsteve should be able to answer your question.