We have an SSAS Multi-dimensional instance that we have been having some issues with for a while. We recently started monitoring it using a tool and noticed that non-shrinkable memory on the instance continues to grow and never seems to clear up. Ultimately, once it passes the total memory limit, we start having issues that require us to restart the instance to get things back working again. This clears everything out the cache and over the course of a couple of weeks it grows back up. The issue requiring a restart may not be related to the memory, but this has been a trend we have seen since we started monitoring.
We have 6 Cubes on this SSAS instance and each one processes nightly. In addition to the cube processing, there is some reporting that runs off of this.
My question is not so much about the issue, but what would cause the non-shrinkable memory to grow like that. My understanding is that non-shrinkable memory is metadata which in our case consists of the objects in 6 cubes which I do not think should amount to much. The active sessions would should clear following there completion. And the Query Memory and Process memory which is used during processing. But my understanding is this should should clear following the jobs running and in our case it is not. It just seems to grow and grow until it exceeds the total memory limit. Almost seems like there is a memory leak somewhere.
The server we are running on has 170GB of memory. The Total Memory limit is set to 85% (roughly 145GB). It takes about 2 weeks of things running before the non-shrinkable memory hits 145GB and shortly after that we start having issues.
Here is what we are seeing via monitoring. The top red line is the hard memory limit, the orange line is the total memory limit.
We are using SQL Server 2012 SP4 Multidimensional SSAS
Any information you can share on why this could be happening or things to check would be much appreciated