Classic vs LakeView Dashboards

Author: Jason Drew

Intro:

LakeView Dashboards are a new offering from Databricks. The goal isn’t to replace what we would call Tier One BI products like Tableau, Power BI or Looker, but rather to make it simpler to Start, Manage, Scale and Distribute data to analysts within the Databricks environment.

The idea behind LakeView is to make it easier for the non-technical business user to easily discover and share insights in a single interface without having to write/save queries and visualizations in one place and construct dashboards in another. Rather LakeView has two tabs, a Data tab for searching for Unity Tables or writing queries that will serve as your Dataset(s), and a Canvas tab where your visualizations are assembled on the fly. This allows a much more Dashboard-centric BI development approach, where users can create dashboard specific data models and deploy those models directly with the Dashboard, instead of having a query-centric model.

I put together a simple dataset consisting of Northwinds database tables. This is a well established classical example database datasoure originally for classical RDMS systems. If you are interested in how I created this LakeView dashboard and the BI development workflow, you can follow the detailed article published here. I created two dashboards, one with Databricks Classic (“Redash”) Dashboards and the other with LakeView. Using the same data and attempting as close to the same output as possible, I came up with 6 evaluation criteria to compare the experience of Lakeview vs Classic:

1. Rendering Speed
2. Layout Flexibility
3. Development Experience
4. Object Reuse
5. Charting Options
6. Unity Integration

Rendering Speed:

LakeView is the clear winner here without question. From clicking on a link to the full rendering of each cell, LakeView appears roughly 3X faster. Prompts also appear to filter near instantly, vs having to select, then apply and finally wait for cells to individually refresh in Classic.

Layout Flexibility:

Resizing of images appears a bit better in Classic. As you can see below, Classic will simply make the object smaller within its widget, while LakeView keeps the aspect ratio constant forcing you to scroll vertically and horizontally when short on space.

It’s nice to have the ability to put Prompts anywhere you like. Classic forces these to the top of the Dashboard whether you want it there or not.

Development Experience:

From scratch, LakeView is much faster. Part of the reason, and why I say from scratch, is that you pick or define your dataset from within the Dashboard from the Data tab.

Once you’ve identified your table or written your query, you simply go back to the Canvas, click the icon to add visualizations, then place and configure them within the Dashboard itself.

You also have the ability to work on Dashboards without having to publish every change during your development process which effects what your viewers will see.

Object Reuse:

Whether you like the process of creating a query, then visualizations then adding those visualizations to your Dashboards or not, Classic gives you the ability to reuse those same visualizations in multiple Dashboards. This is simply not an option currently with LakeView. That said, you could clone a Dashboard which would let you reuse the Data and Visualizations, then make changes to create a new one.

Since it’s pretty quick to make new Visualizations within LakeView, I can’t say that I really miss the Classic Dashboard’s reuse ability.

Charting Options:

From a pure number of charts perspective, Classic has many more than LakeView (23 vs 8 at the time of this writing), but that is bound to change over time.

Classic currently also has more formatting options for things like numbers.

However, one thing that is HUGE going for LakeView is its ability to set the aggregation level on a Counter. This not only saves you from having to create a different query just for your KPIs, but also allows you to apply the same filters to your KPIs which you can’t do in classic if you’re doing anything more than counting rows.

Since the aggregation queries don’t have a reference to Categories, it can’t filter the Counters properly in Classic. Unfulfilled Orders does work as it’s simply a count which is all Classic can do without specifying the aggregation in the backing query. Since we can specify the aggregation for the Counters in LakeView they work seamlessly.

Unity Integration:

When defining your Data set in LakeView, you can only search for Unity tables.

That said, if you make your Dataset out of SQL, you can specify the 3 level namespace to a Hive table.

When you use Unity tables, you get extra capabilities such as the ability to track Lineage, Insights and Quality of your tables from the Catalog Explorer.

From LakeView dashboards, as a viewer, you can also easily see the tables used by clicking the Data Lineage icon in the upper right hand corner.

Wrap Up:

LakeView is off to a nice start. I like the theme of ease of use and I know more Visualizations and potentially even a semantic layer are coming in future releases. Two things I’d also like to see would be drill down and drill through capabilities.

High Level Comparison: