calindamian / Executing-R-script-inside-SQL-Server-

Alternative of SQL Server sp_execute_external_script to allow mixing of SQL and R language

Executing-R-script-inside-SQL-Server-

Alternative of SQL Server sp_execute_external_script to allow mixing of SQL and R language

Objective

Have you allready got frustrated of Microsoft sp_execute_external_script (aka Machine Learning Services) and limited support of allowing setting any version of R language ?

If the answer is yes this post is for you...

Usually you have to wait for a new version of SQL Server that hopefully supports the latest version of R.

I 've created a stored procedure ( based on the old xp_cmdshell ) that allows running R scripts in whatever version you like.

First I explained how to install it and then how you can use it.

Installation

1. Download and install an R language runtime version on your SQL Server machine from https://cran.r-project.org/bin/windows/

2. Once you've installed R make sure you install the following packages:

   • install.packages("tidyverse")
   • install.packages("DBI")

3. Open an SQL Client and open install_execute_rscript.sql.

   Modify the following variables to match your environement

   • @rscriptexe : path to Rscript.exe (eg: 'C:"Program Files"\R\R-4.1.2\bin\Rscript.exe' ) The login used for SQL Server should be allowed execution.
   • @rscriptfile: path to main script that wraps R SQL execution (eg: 'D:\Temp\execute_external_rscript.R' )
   • @output_folder: name of temporary folder used to dump temp files used by the script. (eg: 'D:\Temp'). The login used for SQL should have access to this folder.
   • @conn_input: connection string to be used for the input query. (eg: 'driver={SQL Server};server=SQLSERVER;database=StackOverflow2010;trusted_connection=true') Make sure the login used for SQL Server has access to the configured SQL Server
   • @conn_output: connection string to be used for storing the table containing output results. (eg: 'driver={SQL Server};server=SQLSERVER;database=TMP;trusted_connection=true') Make sure the login used for SQL Server has access to the configured SQL Server

4. Execute install_execute_rscript.sql.

5. Copy execute_external_rscript.R in configured @rscriptfile path.

Usage

The execute_rscript stored procedure has the following params:

  @sql:                      input sql query representing input dataset.
  
  @rscript:                  R script to be executed in the current context.
                                eg:  library (lubridate) ;
                                      .tbinput %>% 
                                       mutate (CurrentDate = today ()) 
                                The input dataset is accessible from R script via **.tbinput** variable.
                                The output of the R script should be in the tibble form so we can output a table back to SQL.
                                
  @output_table:              Output table name to store the output dataset return from the script.
                                      - the table name used to store the output dataset
                                      - If NULL the script will be executed but the output wont be stored as table
                                      - If'AUTO' the table name is uniquely generated and the table name is retured in @output_table variable
                                      
  @drop_output_table:         Controls the persistance of the output table (default 1)
                                       - If  @drop_output_table = 1 output table is droped.
                                       - If  @drop_output_table = 0 output table is kept.
                                       
  @script_output              Controls additional script shell output (default 0) 
                                       - If  @script_output = 1 shows additional shell output.
                                       - If  @script_output = 0 suppress shell output.          

Exemples

---- input data and append an R calculated column
exec master..execute_rscript @sql = 'select  cast ([ClosedDate] as date ) as  [ClosedDate] , 
									count (*) as Cnt ,
									max (Score) as Score ,
									sum (cast (ViewCount as bigint ) ) as ViewCount 
									from [StackOverflow2010].[dbo].[Posts]
									group by cast ([ClosedDate] as date )',

								@rscript  = 'library (lubridate) ;
											.tbinput %>% 
											 mutate (CurrentDate = today ()) ' ,
								@output_table  = 'tmp_out'



---in order to type the results use WITH RESULT SETS
exec master..execute_rscript @sql = 'select  cast ([ClosedDate] as date ) as  [ClosedDate] , 
									count (*) as Cnt ,
									max (Score) as Score ,
									sum (cast (ViewCount as bigint ) ) as ViewCount 
									from [StackOverflow2010].[dbo].[Posts]
									group by cast ([ClosedDate] as date )',

								@rscript  = '.tbinput %>% 
											  summarise(correlation = cor(ViewCount , Score)) ' ,
								@output_table  = 'tmp_out'

WITH RESULT SETS (  (correlation float  ));  

--- execute as script with no table output and activate script output
exec master..execute_rscript	@rscript  = 'print (sessionInfo () )' ,
								@script_output = 1

Conclusions

Wait for your comments and suggestions and if you find it useful the crypto donations are more than welcomed ...

BTC wallet : bc1qm06aatgwn8ducjtexu6cf0wle0d9r00fy7yp7y

or via paystring : calin.damian.crypto$paystring.crypto.com