Archive for September, 2016

ODI 12c new features: Dimension and Cubes! Part 2 (Loading using Natural Keys)

Posted in Cubes, Dimensions, ETL, New Features, ODI 12c, ODI Architecture with tags , , , , , on September 14, 2016 by radk00

Hi all, let’s continue with our posts regarding “ODI 12c new features: Dimension and Cubes”. As stated in the previous post, we can have two ways to build our new objects: with natural keys or with surrogate keys. Today’s post will focus on loading the dimensions and fact tables that where created using natural keys (please see our previous post for all the settings required for those objects).

Let’s begin loading our TIME dimension (which was mapped to our TIME Oracle table). This dimension will have information from three different source tables: SRC_YEAR, SRC_QUARTER and SRC_MONTH. Each of them has information regarding each TIME hierarchy level, so all of them needs to be loaded in order to have a complete hierarchy in our final table.

The load process is very easy and intuitive: first create a new mapping and drag and drop the TIME dimension to it. Then, just add the three source tables, map to its correspondent level in the TIME dimension and that’s it. A very cool thing here is that ODI understands each level as a “separate” table/process, so you don’t need to join your source tables before actually loading it to the target dimension. In other words, ODI allows you to have any kind of complex ETL to each dimension level and each level will be treated as “separate” data loads that will be glued together by the hierarchy setting that you mapped in the TIME dimension object. Here is what it looks like:

blog1

blog2

blog3

blog4

When you execute the mapping we are going to see that the first “MAP_BEGIN” section will try to create and truncate our stage tables that were set in our dimension object. Here is an odd thing (as we also mentioned in the last post): We could not understand yet why ODI “forces” you to have the stage tables created prior to execution (so you can select them in the Dimension object), as it could very well create them for you (like it does for C$ and I$ tables). I know that Oracle may had a reason behind it, but as for now, the entire “stage tables” thing seems an unnecessary setup. Anyway, the important thing here is that ODI will truncate the stage tables before any new execution.

blog5

In the “MAP_MAIN” section is where it gets interesting. We can see here how ODI threats this new dimension object: each level has its own ETL, as we can see that it is loading YEAR, QUARTER and MONTH separately. First YEAR step will load its source to its stage table STG_YEAR, then QUARTER step will join the information from its source table plus STG_YEAR to its STG_QUARTER table. Finally, MONTH step, that is our leaf/grain level, will join its source table plus STG_QUARTER table (which is already joined with YEAR source) and merge it all together in our final table TIME. The result will look like below:

blog6

Since we are not using Surrogate keys here, our Dimension table will contain only the grain/leaf members with all Natural Keys and its attributes for all levels that exists in the dimension. So one row will contain all information regarding all levels that it belongs to. When we create the mappings for the other two dimensions (they’re very similar, so I’m not adding them here) and execute them, we will get the following results:

blog7

blog8

Let’s to go our Fact table load. This one is way too simple, since our source table already contains all the Natural Keys that will be the ones that will also exist in our FACT table (remember, we are not dealing with Surrogate Keys in this example). Here we just need to map each NK to its respective dimension column and also our Measure data and execute the mapping.

blog9

blog10

When we take a look in Operator, we are going to see a single merge command in our Fact table, where ODI will use all dimensions to search if that row already exists in our FACT table. If it exists, the measure column is update, otherwise it is inserted.

blog11

The final result is below: as expected, all Natural Keys from our dimensions were inserted in the Fact table, together with our measure.

blog12

Now you may be wondering, why should I use these new features if it seems a lot of work (settings) for a little gain? Well, using ODI for Natural Key’s only is really not worth it, since the only benefit here seems to be ODI loading the dimensions levels all at once, with different sources/ETL, in a single mapping object, which is a very cool feature, since it enables us to better organize our DW objects and have a clear view on our ETL logic. But again, this is too little for the amount of work that we need to do to get there. But don’t worry, it will get way better when we start to work with Surrogate Keys, since ODI will be able to abstract all the Surrogate Key management and you will start to feel that all the necessary settings will finally be worth the work.

That’s it for today folks! We will be releasing the Surrogate Key settings and load posts very soon, so stay tuned in our blog! See ya!