Hello! Last week I presented for the first time at a SQL User Group (SQL Supper.) There was a lot of content, and plenty of demos, and a few people came up to me afterwards and asked for the scripts. In case you were too shy to ask, or just want a simple database that uses partitioning and Clustered Columnstore Indexes, the scripts attached will provide you with such a thing.
Recently the SQLCAT team published a post on their blog titled Data Loading performance considerations with Clustered Columnstore indexes. In it they talk of loading data into partitioned tables where the load is aligned. Interestingly, their recommended process is to load into a staging table with the clustered columnstore index created first.
This is still based on the fact that you can hit the magic number of 1,048,576, or at the very least 102,400, per bulk insert, to create compressed segments. If you cannot, then I still feel that the method of loading into a heap, and then creating the clustered columnstore index, is preferential. This is because if you trickle insert into a CCI, you will end up with a mixture of open/closed delta stores and come compressed segments. You could leave it to the tuple mover to compress these, and you would still be left with some open delta stores. However, as Remus Rusanu points out, this is not recommended. To complete the process quickly, and to remove any open delta stores, you would need to rebuild the partition.
I was catching up on some blogs yesterday, and one I really liked was this post on Spaghetti DBA. The authors main complaint was that he had grown disillusioned with Connect and the way in which so many issues are dealt with (ie, closed as “Won’t” Fix"). All the author wanted was better feedback from the engineers when responding to issues. I’m adding my voice to this request because in the past year I have raised a couple of issues myself, only to be disappointed with the feedback:
An item has very recently been published to the Microsoft Connect website with regards to an issue relating to Clustered Columnstore Indexes. Unexpected behaviour when using columnstore index | Microsoft Connect. Put simply there is inconsistent behaviour when returning the data on a table that is stored using a Clustered Columnstore Index and a table that is not. There is also an issue when using the “Include Actual Execution Plan” from Managment Studio.
If you’re using, or at least read up on Clustered Columnstore Indexes then you’ll know that the data is stored in segments. You’ll also know that there are 4 states the segments can be in at any one time: