How to Refresh Data Models in Office 365 from On Premises Using Nintex Workflow – A Hybrid Approach

One of the viable use cases for Power BI is to provide reporting to a mobile workforce, even when the organization has invested in BI on premises. In this scenario, there may be a SharePoint farm using PowerPivot for SharePoint with multiple workbooks connected to a myriad of data sources, including SQL Server Analysis Services cubes and models.

The Problem

In this scenario, the workbooks required by the mobile users can be copied up to the Office 365 tenant, and consumed from there. However, how is the data kept current? One way to do so is to set up the Power BI Data Management Gateway (DMG) to do this. The DMG is a great solution, but is relatively new, and has a few inherent limitations at the moment. Chief among them is that data can only be refreshed from SQL Server or Oracle data sources. For the moment at least, Analysis Services is left out in the cold. However, PowerPivot for SharePoint doesn’t share these limitations, and has no problem refreshing data from a wide variety of sources. Unfortunately, if you’re using Power BI, and need to refresh from an unsupported data source, you’re out of luck for the moment, so this is a problem.

Keeping a cloud copy of a local workbook carries the inherent problems of managing two different copies of the same workbooks. Every modification must be performed twice and there is always the chance that something will get missed. Ideally, for this scenario, we should have a publishing mechanism. We can use Nintex Workflow (on premises) to provide this publishing mechanism for us. Whenever the model is refreshed, the workflow will fire, and copy the file up to Office 365 to update the model there.

In order to implement this solution, you’ll need a copy of Nintex Workflow. The low end (Team) edition is fine, but what we’re after is the “Copy document to Office 365” action, and that is available through the Nintex Live action set.

The Solution

To start, we’ll work with an Excel workbook that is using an embedded model created from the standard Contoso Data Warehouse. In this example, we’ve built a simple pivot table to interact with the model. The model has been published to a PowerPivot gallery, and the refresh schedule has been set. Once we’ve tested the refresh to ensure that it’s working, we can create the workflow. To do so, we go to the workflow options section in the library ribbon and create a new Nintex workflow.

We then select the Blank template. Once presented with the design canvas, we select Workflow Settings, give it a good name, and set the workflow to run on create and on edit.

Once the settings have been saved, we can start to build the actual workflow. The first action will be a “Pause for” action. Find it in the Workflow Actions toolbar, and drag it onto the design surface. Select configure from its drop down menu, and set it to pause for a period of time, in this case, 5 minutes.

The amount of time that we need to pause for depends on how long it takes our model to refresh. When the refresh operation begins, it “edits” the document, which kicks off the workflow. We don’t want to upload the workbook until the refresh operation is complete, so we need to pause it. In this case, we are working with a data model that is about 100 MB and comprised of about 3 million rows of data. In this environment, it requires about 2.5 minutes to refresh, so a 5 minute pause is sufficient.

Next, we need to use the “Office 365 upload file” action. This action is one of the Nintex Live actions that can be added at no extra charge to the Nintex environment. If it hasn’t already been added to the toolbar, we need to do so. Assuming that Nintex Live has been enabled for the farm, we browse the available actions through the Catalog which is available through the designer ribbon.

Once in the catalog, we can browse the available actions, and click the Add button to add them to the toolbox. The action that we need to add is named “Office 365 upload file”. Once at least one action is added, an new section will appear in the Workflow Actions toolbar named “Nintex Live”. We then open that section and drag the Office 365 file upload action onto the design surface after the pause action, then configure it.

Walking through the options:

• File to upload: We want the current file to be uploaded to Office 365, so we select current item
• Destination site URL: This is in fact the URL of the site itself. Everything up to, but not including the document library
• Folder path: This is the name of the destination document library, and if appropriate, any subfolders
• File name: We can create or derive a new name for the file, but in this case, we will be using the same name as the source file
• Overwrite existing file: We select this option, as we will be updating existing content.
• Fields: We can add additional metadata values if we wish, but in this case, we leave it empty
• SharePoint Online URL: this is the URL to the root of the Office 365 tenant for the destination
• Username and password are for a proxy account that will be used to upload the file. Secure workflow constants can be used.

Once we configure the action, we save it, our workflow should appear as follows:

and we are ready to publish the workflow. To do so, select the Publish button, then close the workflow designer.

We are now ready to test our workflow. We can force the workflow to run on demand for our workbook, but a better test is to let the refresh trigger it. The most frequently that PowerPivot for SharePoint refreshes can be scheduled to run is once per day. but there is a way to force the refresh to happen on demand.

To force a refresh, we need to edit the refresh schedule for the workbook. Open up the refresh history screen by selecting Manage Data Refresh for the workbook. 

From the Refresh history screen, select the Configure Schedule link to edit the schedule. This will allow us to edit the existing schedule, but more importantly, it will also let us force a refresh by selecting the “Also refresh as soon as possible” option.

If we are using a specific account as a credential, we will also need to re-enter that. Once complete, selecting OK will queue the refresh job for execution. The refresh job will happen the next time that the “PowerPivot Data Refresh Timer Job” executes, and it runs every 5 minutes.

If you’re really impatient, and you have access to central administration, you can edit this timer job and force it to run immediately on a case by case basis.

Moving back to the refresh history screen we will notice that the refresh is currently running.

And in the workflow history screen for the workbook, we can see that the workflow is also running.

Once both operations have completed, we can navigate to our Office 365 site, and we will find the file in the destination folder. If our model is greater than 10 MB in size, we will require Power BI in order to interact with it in the browser, and we will need to enable it for use with Power BI through the Power BI application. If it has already been enabled, then no additional actions are required. If the workbook is less than 10MB, it will work directly in any Office 365 library that is enabled for Excel Services.

More Possibilities

If you’ve worked with the Power BI Data Management Gateway at all, you’ll know that data refreshes can be scheduled at most daily. This limitation is also shared by PowerPivot for SharePoint, but it is possible to change its behaviour. Using the techniques outlined in this article by Ian Smith, and this article by Tim Laqua, refreshes can be made as granular as 5 minutes. These articles are written for SharePoint 2010, but the principles work with 2013 (hint, the stored procedure that forces a refresh is [DataRefresh].[AddRefreshSchedule]). Be warned, this technique is a hack, and not supported by Microsoft. However, it can be a way to up the frequency of refreshes on premises, and with the above technique, in Office 365 too.

Another way to refresh the data more often is by using SQL Server Integration Services (SSIS). SSIS can update an embedded data model in an Excel workbook through a script action (thanks to Gobán Saor).  SSIS packages can be scheduled with SQL agent jobs, which means that they can be run as frequently as every minute. Another advantage of this approach is that with Excel 2013, this method should be fully supported.

Finally…

By relying on our workflow to publish the refreshed model, we work around the current limitations of the Data Management Gateway. We do incur the cost of a Nintex license, but a few days of development would easily exceed that cost. In fact, if our data model doesn’t exceed 10 MB, we don’t even need Power BI for this solution to work.