Extracting raw data from a PowerPivot model using Python

Question

What seemed like a trivial task turned into a real nightmare when I had to read in some data from a PowerPivot model using Python. I believe I've researched this very well over the last couple of days but now I hit a brick wall and would appreciate some help from the Python/SSAS/ADO community.

Basically, all I want to do is programmatically access raw data stored in PowerPivot models - my idea was to connect to the underlying PowerPivot (i.e. MS Analysis Services) engine via one of the methods listed below, list the tables contained in the model, then extract the raw data from each table using a simple DAX query (something like EVALUATE (table_name)). Easy peasy, right? Well, maybe not.

0. Some Background Information

As you can see, I've tried several different approaches. I'll try to document everything as carefully as possible so that those uninitiated in PowerPivot functionality will have a good idea of what I'd like to do.

First of all, some background on programmatic access to Analysis Services engine (it says 2005 SQL Server, but all of it ought to still be applicable): SQL Server Data Mining Programmability and Data providers used for Analysis Services connections.

The sample Excel/PowerPivot file I'll be using in the example below can be found here: Microsoft PowerPivot for Excel 2010 and PowerPivot in Excel 2013 Samples.

Also, note that I'm using Excel 2010, so some of my code is version-specific. E.g. wb.Connections["PowerPivot Data"].OLEDBConnection.ADOConnection should be wb.Model.DataModelConnection.ModelConnection.ADOConnection if you're using Excel 2013.

The connection string I'll be using throughout this question is based on the information found here: Connect to PowerPivot engine with C#. Additionally, some of the methods apparently require some sort of initialization of the PowerPivot model prior to data retrieval. See here: Automating PowerPivot Refresh operation from VBA.

Finally, here's a couple of links showing that this should be achievable (note however, that these links mainly refer to C#, not Python):

• Made connection to PowerPivot DataModel, how can I fill a dataset with it?
• Connecting to PowerPivot with C#
• 2013 C# connection to PowerPivot DataModel
• Connecting Tableau and PowerPivot. It just works. (showing that external apps can in fact read PowerPivot model data - note that the Tableau add-in installs Interop.ADODB.dll assembly, which I guess is what it uses to access the PowerPivot data)

1. Using ADOMD

import clr
clr.AddReference("Microsoft.AnalysisServices.AdomdClient")
import Microsoft.AnalysisServices.AdomdClient as ADOMD
ConnString = "Provider=MSOLAP;Data Source=$Embedded$;Locale Identifier=1033;
             Location=H:\\PowerPivotTutorialSample.xlsx;SQLQueryMode=DataKeys"

Connection = ADOMD.AdomdConnection(ConnString)
Connection.Open()

Here, it appears the problem is that the PowerPivot model has not been initialized:

AdomdConnectionException: A connection cannot be made. Ensure that the server is running.

2. Using AMO

import clr
clr.AddReference("Microsoft.AnalysisServices")
import Microsoft.AnalysisServices as AMO
ConnString = "Provider=MSOLAP;Data Source=$Embedded$;Locale Identifier=1033;
             Location=H:\\PowerPivotTutorialSample.xlsx;SQLQueryMode=DataKeys"

Connection = AMO.Server()
Connection.Connect(ConnString)

Same story, "the server is not running":

ConnectionException: A connection cannot be made. Ensure that the server is running.

Note that AMO is technically not used for querying data, but I included it as one of the potential ways of connecting to the PowerPivot model.

3. Using ADO.NET

import clr
clr.AddReference("System.Data")
import System.Data.OleDb as ADONET
ConnString = "Provider=MSOLAP;Data Source=$Embedded$;Locale Identifier=1033;
             Location=H:\\PowerPivotTutorialSample.xlsx;SQLQueryMode=DataKeys"

Connection = ADONET.OleDbConnection()
Connection.ConnectionString = ConnString
Connection.Open()

This is similar to What's the simplest way to access mssql with python or ironpython?. Unfortunately, this also doesn't work:

OleDbException: OLE DB error: OLE DB or ODBC error: The following system error occurred:
The requested name is valid, but no data of the requested type was found.

4. Using ADO via adodbapi module

import adodbapi
ConnString = "Provider=MSOLAP;Data Source=$Embedded$;Locale Identifier=1033;
             Location=H:\\PowerPivotTutorialSample.xlsx;SQLQueryMode=DataKeys"

Connection = adodbapi.connect(ConnString)

Similar to Opposite Workings of OLEDB/ODBC between Python and MS Access VBA. The error I get is:

OperationalError: (com_error(-2147352567, 'Exception occurred.', (0, u'Microsoft OLE DB
Provider for SQL Server 2012 Analysis Services.', u'OLE DB error: OLE DB or ODBC error: The
following system error occurred:  The requested name is valid, but no data of the requested
type was found...

This is basically the same problem as with ADO.NET above.

5. Using ADO via Excel/win32com module

from win32com.client import Dispatch
Xlfile = "H:\\PowerPivotTutorialSample.xlsx"
XlApp = Dispatch("Excel.Application")
Workbook = XlApp.Workbooks.Open(Xlfile)
Workbook.Connections["PowerPivot Data"].Refresh()
Connection = Workbook.Connections["PowerPivot Data"].OLEDBConnection.ADOConnection
Recordset = Dispatch('ADODB.Recordset')

Query = "EVALUATE(dbo_DimDate)" #sample DAX query
Recordset.Open(Query, Connection)

The idea for this approach came from this blog post that uses VBA: Export a table or DAX query from Power Pivot to CSV using VBA. Note that this approach uses an explicit Refresh command that initializes the model (i.e. "server"). Here's the error message:

com_error: (-2147352567, 'Exception occurred.', (0, u'ADODB.Recordset', u'Arguments are of
the wrong type, are out of acceptable range, or are in conflict with one another.',
u'C:\\Windows\\HELP\\ADO270.CHM', 1240641, -2146825287), None)

It appears, however, that the ADO connection has been established:

• type(Connection) returns instance
• print(Connection) returns Provider=MSOLAP.5;Persist Security Info=True;Initial Catalog=Microsoft_SQLServer_AnalysisServices;Data Source=$Embedded$;MDX Compatibility=1;Safety Options=2;ConnectTo=11.0;MDX Missing Member Mode=Error;Subqueries=2;Optimize Response=3;Cell Error Mode=TextValue

It seems the problem lies in the creation of the ADODB.Recordset object.

6. Using ADO via Excel/win32com, direct use of ADODB.Connection

from win32com.client import Dispatch
ConnString = "Provider=MSOLAP;Data Source=$Embedded$;Locale Identifier=1033;
             Location=H:\\PowerPivotTutorialSample.xlsx;SQLQueryMode=DataKeys"

Connection = Dispatch('ADODB.Connection')
Connection.Open(ConnString)

Similar to Connection to Access from Python [duplicate] and Query access using ADO in Win32 platform (Python recipe). Unfortunately, the error Python spits out is the same as in the two examples above:

com_error: (-2147352567, 'Exception occurred.', (0, u'Microsoft OLE DB Provider for SQL
Server 2012 Analysis Services.', u'OLE DB error: OLE DB or ODBC error: The following system
error occurred:  The requested name is valid, but no data of the requested type was found.
..', None, 0, -2147467259), None)

7. Using ADO via Excel/win32com, direct use of ADODB.Connection plus model refresh

from win32com.client import Dispatch
Xlfile = "H:\\PowerPivotTutorialSample.xlsx"
XlApp = Dispatch("Excel.Application")
Workbook = XlApp.Workbooks.Open(Xlfile)
Workbook.Connections["PowerPivot Data"].Refresh()
ConnStringInternal = "Provider=MSOLAP.5;Persist Security Info=True;Initial Catalog=
                     Microsoft_SQLServer_AnalysisServices;Data Source=$Embedded$;MDX
                     Compatibility=1;Safety Options=2;ConnectTo=11.0;MDX Missing Member
                     Mode=Error;Optimize Response=3;Cell Error Mode=TextValue"

Connection = Dispatch('ADODB.Connection')
Connection.Open(ConnStringInternal)

I was hoping I could initialize an instance of Excel, then initialize the PowerPivot model, and then create a connection using the internal connection string Excel uses for embedded PowerPivot data (similar to How do you copy the powerpivot data into the excel workbook as a table? - note that the connection string is different from the one I've used elsewhere). Unfortunately, this doesn't work and my guess is that Python starts the ADODB.Connection process in a separate instance (as I get the same error message when I execute the last three rows without first initializing Excel, etc.):

com_error: (-2147352567, 'Exception occurred.', (0, u'Microsoft OLE DB Provider for SQL
Server 2012 Analysis Services.', u'Either the user, ****** (masked), does not have access
to the Microsoft_SQLServer_AnalysisServices database, or the database does not exist.',
None, 0, -2147467259), None)

Answer 1

I got in touch with Tom Gleeson (aka Gobán Saor) who was kind enough to let me post his emails here. There are some interesting nuggets in them, so hopefully others will also find them useful.

Email #1

When you say Python, you mean running Python.NET as a standalone exe? If that’s the case, you’re out of luck with Excel PP models (different story for Power BI desktop though). I’ve accessed PP models (2010+) successfully from both VBA, and from Python.NET (via AMO) using similar code to that in your SO question. The difference being (in both VBA & .NET version) is that my code is running in-process within Excel using Excel’s various add-in technologies. (Likely Tableau is also running as an add-in or has embedded Excel within itself enabling similar behaviour). DAX Studio (a useful C# code base to learn the how-tos of PP access) runs both as an Excel add-in and as a standalone EXE, but only as an add-in can it access Excel based PP models.

Email #2

You might find the process of using Python.NET for this somewhat challenging. You would need to embed a Python engine using C#/VB.NET Excel add-in code. I’ve used Excel-DNA (a fantastic open source project) rather than MS’s highly cumbersome "official" method for developing such .NET addins in the past, but I mainly stick to VBA where at all possible.

Using VBA you’ll not be able to access the .NET-only AMO (so no ability to create calculated columns on the fly), but by loading the resulting dataset into an ADO recordset you should be able to output to a worksheet OR to a corporate-database/MS Access OR to a flat-file/CSV etc.

Unlike the 1M worksheet limit, for a flat-file or database output memory (RAM) will be the limiting factor, but, assuming you’re using 64bit Excel and have enough memory to hold the compacted model and the workspace for the largest of the model’s tables in un-compacted form (i.e. a row based rather than column based format that’ll result from a DAX Query), multiplied by 2ish (one instance within PP workspace the other within VBA’s ADO workspace) you should be okay.

Having said that, I’ve never attempted extracting a very large dataset, and using models as a dataset exchange medium is not one of PP’s "use-cases"; so, very large tables might hit some other bug/constraint!

Answer 2

The problem with getting data out of PowerPivot is that the tabular engine in PowerPivot runs in-process inside Excel and the only way to connect to that engine is to have your code running inside Excel too. (I suspect that it may use shared memory or some other transport, but it's definitely not listening on a TCP port or a named pipe or anything like that which would allow an external process to connect)

We do this in Dax Studio by running a C# VSTO Excel add-in in Excel. However that was only designed to work for testing analytic queries, not for doing bulk data extraction. We marshal the data across from the add-in to the UI using a string variable so the entire dataset must be less than 2Gb or the response gets truncated and you will see an "unrecognizable response" error (the data is serialized into an XMLA rowset which is quite verbose so may see it break when only extracting a few hundred Mb of data)

If you wanted to build a script to automate extracting all the raw data from a model I don't think you will be able to do it with Python as I don't believe you can get the python interpreter running in-process inside Excel. I would look at using a vba macro like this one http://www.powerpivotblog.nl/export-a-table-or-dax-query-from-power-pivot-to-csv-using-vba/

You should find that you can query the model for a list of tables with something like "SELECT * FROM $SYSTEM.DBSCHEMA_TABLES" - you could then loop over each table and extract with a variation of the code in the above link.