Great video! Service Principal needs to get granted with the permissions on the system catalog. Its something doesn't come out of the box currently with the template based app, afaik.
Since pricing doesn't really scale down with the complexity of the app, there's no chance I'd deploy a dozen tiny apps. However, if you have an app with a bit of complexity/capability the cost might be justified.
You'd assumably bundle a bunch of different functions into a single "app" with some quick tabs/links, that's all easy enough. Only question would be around scale if that little 2 core box gets busy
I had assumed it worked like the serverless compute, where it only spins up when someone runs it. i.e. its off by default (or scaled to 0), then when someone opens the url it would then turn on the compute and the app. That would make a lot of sense. However from your video it looks like that's not the case, and it's on all the time, or at least you are charged all the time. Is that right? It doesn't really then benefit from serverless compute.
Not really - You've never been able to run python directly inside PowerApps that I'm aware of. You can create super sophisticated workflows via PowerApps, a lot more than you can do here. But if you've got a data engineer who knows python, they can whip up an easy interface without having to build out the integrations & learn the dark arts of the PowerApp expressions language
🤔Is this kind of a retro thing? Terrible, expensive, sluggish ... all the attributes we are looking for in (almost) 2025. If you think a self-hosted container is the alternative to this, you are either living under a stone, or you're well paid by Databricks (maybe).
You forgot slow... But you seem to be a bit mistaken. I meet many customers who ask for just that on a daily basis. Databricks does not push features for the sake of it, the demand is there and it fits specific needs for quick prototyping, leveraging popular lightweight app servers. Streamlit itself is terrible, expensive, sluggish, and... popular, worth $800M when it was aquired 3 years ago by snowflake. So... 🤷🏻
Interesting to see whether existing lightweight apps are migrated/re-developed in Databricks Apps and what the migration path is!
Great video! Service Principal needs to get granted with the permissions on the system catalog. Its something doesn't come out of the box currently with the template based app, afaik.
Since pricing doesn't really scale down with the complexity of the app, there's no chance I'd deploy a dozen tiny apps. However, if you have an app with a bit of complexity/capability the cost might be justified.
You'd assumably bundle a bunch of different functions into a single "app" with some quick tabs/links, that's all easy enough. Only question would be around scale if that little 2 core box gets busy
I had assumed it worked like the serverless compute, where it only spins up when someone runs it. i.e. its off by default (or scaled to 0), then when someone opens the url it would then turn on the compute and the app. That would make a lot of sense. However from your video it looks like that's not the case, and it's on all the time, or at least you are charged all the time. Is that right? It doesn't really then benefit from serverless compute.
atuhenticate is pat token? we have issues where the databricks cannot be install
What does deployment to test and prod look like? Is asset bundles involved?
They are creating .py and .yaml files, I believe it would be easy to deploy using asset Bundles. Similar to the workflow yamls.
Ahh, ok. So PowerApps, but from 5ish years ago. Ok. Let's see how it evolves.
Not really - You've never been able to run python directly inside PowerApps that I'm aware of. You can create super sophisticated workflows via PowerApps, a lot more than you can do here. But if you've got a data engineer who knows python, they can whip up an easy interface without having to build out the integrations & learn the dark arts of the PowerApp expressions language
I'll get interested when we can write back to the lakehouse.
It'd be useful for mappings and config stuff that's currently hard coded or in a SQL DB.
🤔Is this kind of a retro thing? Terrible, expensive, sluggish ... all the attributes we are looking for in (almost) 2025. If you think a self-hosted container is the alternative to this, you are either living under a stone, or you're well paid by Databricks (maybe).
You forgot slow... But you seem to be a bit mistaken. I meet many customers who ask for just that on a daily basis. Databricks does not push features for the sake of it, the demand is there and it fits specific needs for quick prototyping, leveraging popular lightweight app servers. Streamlit itself is terrible, expensive, sluggish, and... popular, worth $800M when it was aquired 3 years ago by snowflake. So... 🤷🏻