Sql-server – Deploying a cube where database name has been changed

I will tell you the way we deploy in my place SSAS projects between different environments. We use a set of Powershell scripts that:

Part 1:

• get the last version from the build server and deploy the file to generate the last working version of XMLA;
• take the generated XMLA from the build server and copy it to the QA server;

Part 2:

• Now, on the QA server, we change manually the server name, the database name (set as parameters to the PS script), we verify that all required scripts are there;
• one PS script runs the resulted XMLA on the current QA server, so the SSAS database is created (not processed though);
• a XMLA script that commands the full process of the cube and its dimensions;

All of these steps are done as automatically as possible, with as little intervention from the developer/QA guy as possible. We do, though, insert some eyeballing in between, just to keep our minds safe. I won't say that it's the best, but it works and gets the job done. Before the last step I'm sure you can insert any XMLA script to remove existing roles and create new ones. And to change a PS script to execute some new SQL or XMLA script is not hard at all. I suppose that the roles and permissions don't change that much, so they can be scripted and reused.

PS: this is an environment with domain users and fixed servers (so we can use shares, domain accounts and all that fluff). If you have another environment it might be a more difficult to get it done.

I found the answer eventually ....

There was a job set up to incrementally process the cube. Incremental cube processing requires a query to be defined to select new fact data to merge into the cube. In my case, this query was defined badly and was selecting the entire set of fact data, i.e. all the data that was already in the cube. Although fully processing the cube did not consume excessive disk space, for some reason incremental processing consumed all the available space. As a result the process never completed successfully. Rebooting the server caused the files created by the unfinished process to be deleted, but then minutes later the processing job would start again.