Access Violation and SSAS 2016 Multidimensional Crashes Solved

Thanks to the fact that sometimes we work on larger and challenging projects, our tools are pushed to its highest performance and therefore, complicated bugs arise.

While in a PoC to migrate a customer from SQL Server 2014 to SQL Server 2016, my SolidQ colleagues Alberto Spelta and Francesco Milano faced a serious SSAS Multidimensional bug causing a stack Access Violation and a consequent crash of the service. They worked hard with MS Support and a SSAS Escalation Engineer to delineate and reproduce the issue, which in the end turned out to be a product bug. Here’s a brief description of the bug directly from the Escalation Engineer, along with a suggested workaround to avoid the problem:

In SSAS 2016, there were some changes made to existing functionality to refactor throttling in MOLAP engine, which was done to fix the issue hang issues reported earlier. Because of the changes, we ran into this bug.

Because of this bug, the code triggers the next partition in the queue of waiting jobs. If lots of partitions are in the scheduled, the stack becomes extremely deep. Our PG had worked on this and fixed.

Until the fix is released as a part of future cumulative update, we request you to use the workaround of setting “CoordinatorExecutionMode” to 500 (in the msmdsrv.ini) in your customer environment.

Now a fix addressing this issue has been just released in the Cumulative Update 3 for SQL Server 2016 SP1 (KB 4019988):

So, if you're working on a big MOLAP database having a lot of partitions in SSAS 2016, the CU3 for the SP1 is highly recommended.