Qlik Community

Qlik Design Blog

All about product and Qlik solutions: scripting, data modeling, visual design, extensions, best practices, etc.

Announcements
Coming your way, the Qlik Data Revolution Virtual Summit. October 27-29. REGISTER

Calculated fields are often created in the script and stored under new aliases. But you can also create them in the user interface. What are the pros and cons with the two methods? And how are the user interface fields calculated?

Normally, new fields are created in the script and stored as additional columns in the data model. Just write your expression inside a Load statement and you’re done:

Calculated field in script.png

But you can also do the same thing in the user interface, and then it could look like this:

Year as dimension.png

 

So, which way should you do it?

Generally, I would say that you should put as much as possible in the script. In most cases, it is far better to have these calculations pre-made, so that they do not have to be calculated at run-time, i.e. when the user clicks.

The Qlik Engine has two fundamentally different ways to calculate such expressions: As “Calculated dimension” or as “Field-on-the-fly”. The engine automatically decides how a specific calculation should be made, depending on the expression.

 

Fields-on-the-fly

This method was introduced in one of the early versions of Qlik Sense. As the expression is evaluated, the engine creates an additional column in the data model, with a corresponding symbol table. Just as for a real field, the selection is stored in state vectors linked to this column.

In the picture below you can see a table dimension defined as “=Year(Date)”, which results in four rows.

Now look at the selection bar: When a selection is made, the corresponding year is selected in the Field-on-the-fly called “=Year(Date)” – a field that does not exist in the original data model. And in the selection bar you can see that the selection is indeed stored in this “virtual” field, and not in the Date field.

Field on the fly.png

 

Calculated dimensions

This is the old-fashioned way, and this is how QlikView still today does it. In the example below, the table dimension is “=Aggr(Year(Date),Date)” and it also results in four rows. Logically, this expression is equivalent to the above one.

But here the selection is instead made in in the underlying field: in the Date field.

Calculated dimension1.png

 

It is always possible to create a Calculated dimension, no matter what the expression looks like. But the same is not true for Fields-on-the-fly. There are limitations to when they can be generated:

  • The expression must be based on one single field only, or on multiple non-key fields from the same table in the data model
  • The expression cannot depend on the selection state, e.g. through the GetSelectedCount() function 
  • The expression cannot contain an Aggr() function

 

If a Field-on-the-fly cannot be generated, the expression will be evaluated as a Calculated dimension instead.

 

Performance

Both Calculated dimensions and Fields-on-the-fly can cause performance problems, so it is a good idea to consider moving them to the script instead. Fields-on-the-fly can almost always be moved to the script.

For Fields-on-the-fly, the performance problems become especially severe if the underlying field has many distinct values. A common example is when calendar functions like Year and Month are used on a timestamp with millions of distinct values, rather than on a date with fewer values, like 2 x 365 dates. Further; since Fields-on-the-fly are added to the data model, and the hash of the data model is used in the ID of the cache entry, Fields-on-the-fly can prevent the cache from being re-used properly.

To improve the performance, Master dimensions containing Fields-on-the-fly are now (since Nov 2019) calculated already when the first user opens the app, something which can increase the time it takes to open a document. On the other hand, this will improve the response time considerably in the analysis phase, as well as mitigate cache problems, so we are confident that this is a correct decision. Hence, put your Fields-on-the-fly in the Master dimensions!

Should you want to tweak the behavior of the engine, you can always try the following:

  • Using “UseAutoFieldOnTheFly=0” in Settings.ini will disable Fields-on-the-fly for all documents served by the engine
  • Using “Set QlikInternalDisableFotfMode=1;” in the script will disable Fields-on-fly in the app
  • Using “Set QlikInternalDisableFotfPregen =1;” in the script will prevent Fields-on-fly from being pre-calculated when the app is opened
  • Wrapping the expression in “=CalcDim(…)” will force it to be a Calculated dimension

 

But most importantly - don't use a timestamp to create your calendar! Use a date instead:

Timestamt2Date.png

Good luck!

HIC

Tags (1)
13 Comments

@mmarchese 

Your question is very relevant. But it only concerns Fields-on-the-fly in the selection bar. Calculated dimensions never appear in the selection bar and labels in other places are manually set - as they should be.

But for Fields-on-the-fly, I agree with you: There is a lot of room for improvement... but it is not as simple as one might think, and there are some considerations to be made:

  1. The same expression may exist in several objects, and thus have different labels. Which label should you then use? If it is a Master item, then it's easy: Master items trump other labels.
  2. The same object dimension can in fact create several selections. Just look at a histogram, where the dimension is a Class() function with variable bin width. Should the different selections use the same label? Then you could get many selections side by side that all have the same label.

But these problems can be solved, and I promise you to push for more development in this area. 

1,070 Views
Creator II
Creator II

@Henric_Cronström Ah yes, the devil's in the details.  I'm glad it's on Qlik's radar though.  Thanks!

0 Likes
1,054 Views
Partner
Partner

Thanks @Henric_Cronström 

A really good  blog post. I learned something new. I've noticed the difference between the two calculated dimension behaviours before but haven't really given it any deep thoughts until now.

I've noticed that front end calculated dimensions is an frequent topic in the Community forums so I will for sure reference your post when needed in the future. 

BR Vegar

988 Views
Luminary
Luminary

@Henric_Cronström thank you for discussing Field-on-the-Fly, its something thats interested me since I saw it got added to the Engine API. I have a few follow up questions on how it works:

  1. You mention that you suggested moving FotFs to master dimensions in Sense so they get created up front. Does that imply then that the performance hit of using them is a 1 time thing? Ie, they impact performance at the time they are created, but once indexed they work like other fields? Or do they cause slower performance in say, a hypercube calculation, as opposed to a persisted field?
  2. Via the Engine API, we can name our FotFs. Let's say I create one with the Class() function and call it [MyCustomBin]:
    1. Are we able to use [MyCustomBin] like a native field in a list object, with search and selections?
    2. Are we able to use [MyCustomBin] in set analysis expressions?
    3. Will [MyCustomBin] be visible anywhere in the TablesAndKeys data that the Engine provides about a data model?
  3. The EngineAPI has "AddFieldFromExpression" for adding a field on the fly. Is there a similar "Remove Field on the Fly" method? Would there be any benefit to dropping these FotFs if they become unused?
0 Likes
753 Views
  1. Is the performance hit a 1-time thing? Yes, in the sense that they are calculated and cached. But there is a theoretical possibility that they are purged from the cache if they are not used often.
    They should not per se cause slower performance in a hypercube calculation.
  2.  I have not used the Engine API to create fields-on-the-fly myself, so I might be wrong... but I think the answers are Yes, Yes, and No. 
  3. I don't think there is a "Remove field on the fly", and I don't think there would be any benefit either.
0 Likes
578 Views
Employee
Employee

I've done some testing on this relative to your first question and the answer in my experience is a solid, NO.   I did very little clicking around and the calculations were dropped out of cache and had to be recalculated; so while HIC correctly identifies this as a theoretical possibility, I found the likelihood of it happening is high.

My testing was done around calendar calculations associated with derived calendars the load script Declare and Derived Field statements.    The derived fields are calculated field definitions that are applied behind the scenes.   Derived calendar fields tend to have very few distinct values yet the calculated field took several seconds to calculated the first time.  After that the performance was effectively identical to a data model based calendar - a few milliseconds.  However, despite the size of the calculated field being very small, it would get dropped out of cache as I moved around the app performing my analysis.

Because of the performance hits I've always avoided calculated fields whenever possible and upon discovering and validating the performance impact of "Derived" fields I've stopped using the Declare-Derive unless some aspect of the analytic requires a dynamic dimension.

Best Practice (to reiterate HIC's statement above): 

"Both Calculated dimensions and Fields-on-the-fly can cause performance problems, so it is a good idea to consider moving them to the script instead. Fields-on-the-fly can almost always be moved to the script."

565 Views