Legend not working correctly in stacked column chart

[charleshale]

The attached PBIX is a simple Revenue By Customer (anonymized).  The data looks like this.  I also have a date table.

When I do a stacked revenue chart without a Legend, the numbers sum correctly (see first tab).

When I do a stacked revenue chart with a Legend, data goes missing and the numbers sum incorrectly.

I've prowled through the support boards and am not finding issues that suggest user error.  Is this a bug?  It's such a simple table with 1:1 relationships that I'm not seeing the user error here.

Test File Stacked Chart - Revenue By Customer

Thank you

Charlie 

[Sergiy]

Hi @charleshale 

The link to the file you provided is not valid.

I'll try to make a guess of what is happening. I think you have either Customer or Date column sorted by another column. If I am right make it sorted by itself and check the result.

Is my guess correct?

[charleshale]

Apologies for the broken link.  Here's it on onedrive:  https://1drv.ms/u/s!AkwttVUD7SbrjCLxQ1M6jf0xeryf

Hmmm - is this doing that?

[Sergiy · Accepted solution]

It seems that Power BI simply can not include/calculate all the customers's data.

If you filter data showing the results for a few customers only then the trend will be OK.

[charleshale]

I agree - seems like a bug, right?   I'd expect the stacked chart to be able to handle N customers up to the memory limits, like excel!

[Sergiy]

Let's just speculate a bit.

How could a chart represent, for example, 1000 customers, if 200px(for example) is available only.

If 1px is for a customer, then for 1000 customers at least 1000px are needed, but what we have is 200px only.

[charleshale]

@SergiyI agree with you - would be silly to try to cram 1000 customers into 200 px.   I like what excel does here:

(i) flip into a clustered column chart instead of a stacked if try a power pivot table, and 

(ii) if you then try to go stacked, it will fail, giving a warning of a 255 max series limit.

FWIW - I think something like the foregoing would be a lot better because, at least for me, it was easy in PBIX to get the chart, not see that I was only capturing half my series, and proudly publish the data to my peers, not realizing that the data series was incomplete!  Darn!  And I was so proud of my work!  So, to me the bug is mostly the alerting.  This said, charting NIRVANA would be a chart that collapses the smallest n customers in the stack.