Snowflake Architecture · Topic 11 of 18
Snowpark Container Services, Streamlit in Snowflake and Native Apps
Choose between Snowpark Container Services, Streamlit in Snowflake and the Snowflake Native App Framework to run custom code, build apps and share them.
- 6 min
- SPCS components: image repository, compute pool, service, job service, service function.
- Compute pool is not a warehouse; egress needs an external access integration.
- Streamlit in Snowflake: Python UI inside Snowflake, owner's rights, runs on an assigned warehouse.
What it is
These three features cover the part of an AI or data application that managed functions cannot. Snowpark Container Services runs your own containers inside Snowflake. Streamlit in Snowflake gives you a Python user interface on top of your data. The Snowflake Native App Framework packages data, logic and interface so another account can install it.
The exam asks you to match a requirement to one of them. The question usually hints at the answer: "custom runtime or GPU" points to containers, "interactive dashboard for analysts" to Streamlit, and "distribute to customers who must not see the code or send us their data" to a Native App.
How it works
Snowpark Container Services
Snowpark Container Services (SPCS) runs OCI container images on Snowflake managed compute, next to your data. Nothing leaves the Snowflake security boundary, and the data stays governed by your roles.
| Component | Role |
|---|---|
| Image repository | Registry for your container images, created with CREATE IMAGE REPOSITORY and pushed to with standard OCI tools such as the Docker CLI |
| Compute pool | A set of virtual machine nodes with a chosen machine type and a minimum and maximum node count; some types have GPUs |
| Service | A long running workload that Snowflake restarts if a container exits |
| Job service | A finite workload that runs to completion and is not restarted, started with EXECUTE JOB SERVICE |
| Service function | Lets SQL send batches of rows to a service |
Compute pools are separate from virtual warehouses. A warehouse runs SQL, while a compute pool runs containers.
Network access is explicit. Inbound traffic to a public endpoint is controlled with Snowflake network policies, and outbound traffic to the internet needs an external access integration. A container cannot call external hosts by default.
Choose containers for a custom model server, a GPU workload, a non Python language, or a third party application that must run next to the data. If a SQL function or a Snowpark stored procedure does the job, do not add containers.
Streamlit in Snowflake
Streamlit in Snowflake lets you build and run Streamlit apps inside Snowflake, written in Python, with no separate web hosting. The app is a Snowflake object (created with CREATE STREAMLIT or in Snowsight), and its code is stored and run in Snowflake.
The security model matters for the exam.
- The app runs with the owner's rights, so viewers see what the owner's role can see.
- Viewers need access to the app object itself, granted to roles like other privileges.
- Queries run on a warehouse that you assign to the app.
- Because owner's rights apply, design the owner role so that it can reach only the data the app should show.
Streamlit fits dashboards, data entry forms, model demos and front ends for Cortex or ML results. It is an interface, not a distribution channel to other accounts. For that, package the app in a Native App.
Snowflake Native App Framework
The Snowflake Native App Framework lets a provider build an application that runs inside a consumer's Snowflake account. The provider ships code and, optionally, data; the consumer installs it and controls what it can access.
| Piece | Purpose |
|---|---|
| Application package | The provider's container for versions, files, a manifest and a setup script, created with CREATE APPLICATION PACKAGE |
| Manifest and setup script | Declare the app's needs and create its objects when installed |
| Application | The installed instance in the consumer account, created with CREATE APPLICATION |
| Listing | How the app is offered: publicly on Snowflake Marketplace or privately to chosen accounts |
| Privileges and references | The app asks the consumer for access to specific objects or account privileges, and the consumer grants them |
Key properties for architects:
- The app runs in the consumer's account, so consumer data does not have to be sent to the provider.
- The provider's logic is protected: consumers cannot read the app's internal code or objects.
- The provider can release new versions and patches, and consumers upgrade the installed app.
- An app can include Streamlit interfaces, stored procedures, UDFs and Snowpark Container Services workloads.
Which one
| Requirement | Choice |
|---|---|
| Custom inference server, GPU training, any language | Snowpark Container Services |
| Interactive dashboard or form for people in your account | Streamlit in Snowflake |
| Sell or share an app with other accounts, code hidden, data stays with the consumer | Snowflake Native App Framework |
| All of them together: a vendor app with a container backend and a Streamlit front end | Native App containing both |
Common mistakes
Calling a compute pool a warehouse. SPCS compute pools run containers and are billed and sized separately from virtual warehouses.
Assuming containers can reach the internet. Outbound access needs an external access integration; inbound access is governed by network policies.
Forgetting Streamlit runs with owner's rights. A viewer can see whatever the app queries on the owner's behalf, even without direct table access.
Using Streamlit to distribute to customers. A Streamlit app lives in one account. Cross account distribution with hidden code is the Native App Framework.
What to remember for the exam
- SPCS components: image repository, compute pool, service, job service, service function.
- Compute pool is not a warehouse; egress needs an external access integration.
- Streamlit in Snowflake: Python UI inside Snowflake, owner's rights, runs on an assigned warehouse.
- Native App Framework: application package built by a provider, installed as an application in the consumer account, shared through a listing.
- Native Apps keep consumer data in the consumer account and protect provider code.
Check yourself
A vendor needs to run a fine tuned open model that requires GPUs next to customer data. Which feature?
Snowpark Container Services, with a compute pool that has GPU nodes and a service that exposes the model.
Analysts need an internal Python dashboard on Snowflake data with no hosting to manage. Which feature?
Streamlit in Snowflake, created as an app object and run on a warehouse.
A provider wants customers to use its analytics without seeing the code or exporting their data. What do they build?
A Snowflake Native App: an application package published in a listing and installed in each consumer account.
Topic 11 of 18 in Snowflake Architecture
Test yourself on Snowflake Architecture
The full Snowflake ARA-C01 question bank has 687 questions, with timed practice exams and an explanation for every answer.
This study guide is independent and is not affiliated with, endorsed by, or sponsored by Snowflake. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.