Posit Team architecture
Overview
The Posit Team Native App manages the installation and lifecycle of Posit products within Snowpark Container Services: Posit Workbench, Posit Connect, and Posit Package Manager. Administrators can choose which products to install.
Administrators install and manage the Posit Team Native App through Snowsight. Posit Team then manages each Posit product and makes it available to your users. The Posit Team Native App runs the controller and each product as services within Snowpark Container Services.
Users can directly reach Posit Team and each product using the specific endpoint for each service. Snowsight also lists links to Posit Team, for users who do not have the URL. See User access.
Components
Posit Team controller
The Posit Team controller is the central management component of the Native App. When you visit Posit Team from Snowsight, you are interacting with the controller. It provides:
- A unified landing page for accessing all Posit products
- Installation and configuration for each product
- Lifecycle management (e.g., start, stop, upgrade) for each product
- Centralized access management through Snowflake application roles
The controller manages each product independently, allowing administrators to start, stop, and configure Workbench, Connect, and Package Manager based on their organization’s needs.
See the Posit Team installation instructions to get started.
Posit Workbench
Posit Workbench is a development environment for data scientists. Within the Posit Team Native App, Workbench provides:
- Positron, RStudio, VS Code, and other popular IDEs for data science
- Direct access to Snowflake data through managed credentials
- Persistent home directories backed by block storage
See Workbench setup for configuration details.
Posit Connect
Posit Connect is a publishing platform for data products created in R and Python. Within the Posit Team Native App, Connect enables:
- Publishing and hosting of Shiny apps, Quarto documents, APIs, and many other artifacts
- Scheduled report generation and distribution
- Secure sharing of data products within your organization
See Connect setup for configuration details.
Posit Package Manager
Posit Package Manager is a repository management server for R and Python packages. Within the Posit Team Native App, Package Manager provides:
- Centralized package repositories for your organization
- Curated package sets with security and compliance controls
- Offline package access within Snowflake
See Package Manager setup for configuration details.
User access
Users access Posit Team products in several ways:
- Via Snowsight: Navigate to the installed app in your Snowflake account and launch the Posit Team landing page
- Posit Team URL: Access Posit Team directly and navigate to each configured product
- Direct URL: Access products directly using the service URLs provided after installation
Snowflake OAuth handles authentication to Posit Team and each product, providing single sign-on for users.
Administration
Administrators use Snowsight to:
- Install the Posit Team Native App from the Snowflake Marketplace
- Manage egress rules for Posit Team and each product
- Create OAuth integrations for use by Workbench and Connect
- Control user access through application roles
Technical details
The Posit Team Native App uses several Snowflake objects to deploy and manage products, including Snowpark Container Services objects such as services and compute pools. This section describes these objects and how Posit Team uses them.
Resource provisioning happens at two stages:
- Posit Team installation: Snowflake creates the controller service, its compute pool, and its warehouse when you install Posit Team from the Snowflake Marketplace.
- Product installation: The controller creates resources such as a per-product service, compute pool, volumes, and secrets when you install Workbench, Connect, or Package Manager.
This means you only pay for the resources associated with the products you choose to install.
Service architectures
Each subsection shows one product and the Snowflake resources it uses. Posit Team manages each product and provides a central way to access all of them. The sections after these describe each type of resource.
Controller
| Component | Role |
|---|---|
POSIT_TEAM (compute pool) |
The compute pool that hosts the controller service. It has one CPU_X64_XS node. Snowflake creates it when you grant the CREATE COMPUTE POOL privilege to the application. |
POSIT_TEAM.SERVICES.CONTROLLER |
The service that hosts the Posit Team controller. Users reach its web endpoint. |
POSIT_TEAM (warehouse) |
The X-Small warehouse the controller uses for queries. Snowflake creates it when you grant the CREATE WAREHOUSE privilege to the application. |
POSIT_TEAM.DATA (schema) |
Holds configuration tables, secrets, and snapshots. |
Workbench with local sessions
By default, Workbench runs a session for each user as a process inside the Workbench service. Maya and Jordan are example users running independent sessions.
| Component | Role |
|---|---|
POSIT_TEAM_WORKBENCH (compute pool) |
The compute pool that hosts the Workbench service. |
POSIT_TEAM.SERVICES.WORKBENCH |
The service that hosts Workbench and all user sessions. Users connect to its web endpoint. |
| Session process | An IDE session for one user, such as RStudio or Positron, running as a process in the Workbench service. |
Block volume /data |
Persistent storage for Workbench application data. |
Block volume /home |
Persistent storage for user home directories. |
Sessions run inside the Workbench service, so they share its CPU and memory. This works well when all users and their work fit on the nodes of the compute pool.
Workbench with external sessions
This feature is in private preview. To enroll, request access from Posit support.
With external sessions, Workbench runs a session for each user as its own job service on a session compute pool. Administrators configure one or more session compute pools, and the number of nodes in each. The diagram shows one pool; session placement shows several. Maya and Jordan are example users, and MY_FILESYSTEM is an example filesystem name.
| Component | Role |
|---|---|
POSIT_TEAM_WORKBENCH (compute pool) |
The compute pool that hosts the Workbench service. It only runs Workbench, so it can use the smallest instance family. |
POSIT_TEAM.SERVICES.WORKBENCH |
The service that hosts Workbench. Users connect to its web endpoint. |
POSIT_TEAM_WB_SESSIONS* (compute pools) |
The compute pools that host session job services. Administrators choose the number of pools, and the instance family and node count of each. The diagram shows one. |
POSIT_TEAM.SERVICES.SESSION_* |
A job service that runs one session. Workbench creates it when the user starts the session and removes it when the session ends. |
POSIT_TEAM.DATA.SESSION_STDIN_* |
A secret that holds the configuration for one session, mounted into the job service for that session. Workbench creates it with the job service and removes it when the session ends. |
POSIT_TEAM.DATA.WORKBENCH_HOME (filesystem volume) |
Holds user home directories. The Workbench service and every session job service mount it. |
MY_FILESYSTEM (filesystem) |
The Snowflake filesystem that backs the filesystem volume. You create it and grant the application access to it. |
Block volume /data |
Persistent storage for Workbench application data. |
Session placement
Each session requests CPU and memory when the user starts it. Two components decide where the session runs:
- Workbench chooses the session compute pool. It picks the smallest pool whose nodes can hold the requested CPU and memory.
- Snowflake chooses the node. It places the session job service on a node in that pool with room for the request, adding a node when needed, up to the pool’s maximum. A session never spans nodes.
The following example has two session compute pools. POSIT_TEAM_WB_SESSIONS has up to four CPU_X64_XS nodes (1 vCPU, 6 GiB each). POSIT_TEAM_WB_SESSIONS_2 has up to three CPU_X64_M nodes (6 vCPU, 28 GiB each). To keep the example simple, every session requests the same amount of CPU, which does not change any placement here, so memory alone decides where each session goes. Maya, Jordan, Alex, Dani, and Kiran are example users.
Both compute pools in the diagram are shown with “available” nodes which represent how the pool might expand.
| User | Request | Pool | Node | Reason |
|---|---|---|---|---|
| Maya | 4 GiB | CPU_X64_XS |
1 | The smallest pool whose nodes hold 4 GiB. |
| Jordan | 5 GiB | CPU_X64_XS |
2 | The same pool, but 4 GiB and 5 GiB together exceed one node’s 6 GiB, so Snowflake uses another node. |
| Alex | 10 GiB | CPU_X64_M |
1 | The request exceeds the 6 GiB nodes of the smaller pool. |
| Dani | 10 GiB | CPU_X64_M |
1 | The session fits beside Alex’s. Together they use 20 of the node’s 28 GiB. |
| Kiran | 24 GiB | CPU_X64_M |
2 | The request fits a node of this pool, but not beside Alex and Dani. |
Workbench rejects requests that no compute pool can satisfy.
Connect
Connect runs published content as processes inside the Connect service. Maya and Jordan are example users publishing independent content.
| Component | Role |
|---|---|
POSIT_TEAM_CONNECT (compute pool) |
The compute pool that hosts the Connect service. |
POSIT_TEAM.SERVICES.CONNECT |
The service that hosts Connect and all running content. Users connect to its web endpoint. |
| Content process | A running piece of content, such as a Shiny application or a Plumber API, running as a process in the Connect service. |
Block volume /data |
Persistent storage for Connect application data, including published content. |
Content runs inside the Connect service, so it shares the CPU and memory of the service.
Package Manager
Package Manager serves R and Python packages and IDE extensions to your users. It gets them by syncing from a Posit service outside Snowflake.
| Component | Role |
|---|---|
POSIT_TEAM_PACKAGE_MANAGER (compute pool) |
The compute pool that hosts the Package Manager service. |
POSIT_TEAM.SERVICES.PACKAGE_MANAGER |
The service that hosts Package Manager. Users connect to its web endpoint. |
Block volume database |
Persistent storage for the Package Manager database. |
Block volume data |
Persistent storage for package archives and metadata. |
| Posit package sync service | A service run by Posit, outside Snowflake. Package Manager syncs CRAN, PyPI, and Bioconductor packages and VS Code extensions from it. It is the one egress destination Package Manager requires. See Egress. |
Workbench and Connect use Package Manager automatically, over the internal network. No user action is needed.
Warehouse
The Posit Team controller runs queries on a warehouse. Snowflake creates the warehouse POSIT_TEAM when you grant the CREATE WAREHOUSE privilege to the application.
| Warehouse | Size | Auto-suspend | Auto-resume |
|---|---|---|---|
POSIT_TEAM |
X-Small | 30 seconds | Yes |
The warehouse uses Snowflake credits only while it runs queries and until it auto-suspends.
Services
Each Posit product runs as a Snowpark Container Services service. A service is the primary runtime entity that executes containerized applications.
Posit Team creates the following services:
| Service | Description |
|---|---|
POSIT_TEAM.SERVICES.CONTROLLER |
The Posit Team management interface |
POSIT_TEAM.SERVICES.WORKBENCH |
Workbench development environment |
POSIT_TEAM.SERVICES.CONNECT |
Connect publishing platform |
POSIT_TEAM.SERVICES.PACKAGE_MANAGER |
Package Manager repository server |
Each service defines containers, network endpoints, volume mounts, and environment configuration. The controller lets you start, stop, and upgrade services.
Compute pools
Compute pools are created under the CREATE COMPUTE POOL privilege that you grant to the application. Snowflake creates the compute pool for the controller, POSIT_TEAM, when you grant it. The controller creates the compute pool for each product when you install the product.
Each product runs on a dedicated compute pool, which provides the compute resources for the service containers.
Posit Team creates separate compute pools for each product:
| Compute Pool | Default Instance Type |
|---|---|
POSIT_TEAM |
CPU_X64_XS |
POSIT_TEAM_WORKBENCH |
CPU_X64_S |
POSIT_TEAM_CONNECT |
CPU_X64_S |
POSIT_TEAM_PACKAGE_MANAGER |
CPU_X64_S |
With external sessions, Workbench also uses session compute pools, which administrators configure.
Administrators can configure the instance type for each compute pool during product setup to match workload requirements. See Snowflake compute pool instance families for available options.
Block volumes
Products use block storage volumes for persistent data that requires filesystem semantics, such as databases and configuration files.
| Product | Volume | Default Size | Purpose |
|---|---|---|---|
| Workbench | home |
128 GiB | User home directories (local sessions only) |
| Workbench | data |
1 GiB | Application data |
| Connect | data |
128 GiB | Application database, published content, and other artifacts of that content |
| Package Manager | database |
50 GiB | Application database |
| Package Manager | data |
500 GiB | Package archives and metadata |
Block volumes persist across service restarts. With external sessions, Workbench stores home directories on a filesystem volume instead of a home block volume. See Snapshots for backups.
Secrets
Products use Snowflake secrets to securely store sensitive configuration such as product licenses and OAuth credentials. Snowflake mounts secrets into service containers as environment variables or files.
Common secrets used by products include:
- Posit product licenses
- OAuth client secrets for Snowflake authentication
- Bootstrap tokens for initial API access
- OpenID Connect (OIDC) client secrets for single sign-on
- Configuration for each external session, one secret per session
Network access
Products require network access both for inbound user connections and outbound connectivity to external services.
Inbound access: Each service exposes HTTP or HTTPS endpoints that provide public ingress URLs for user access. Users connect to Workbench at its endpoint. Workbench proxies the traffic to each session, and users cannot reach a session directly.
Outbound access: Products use external access integrations with network rules to reach external services such as package repositories (CRAN, PyPI, and Bioconductor) and Snowflake OAuth endpoints.
The application cannot create these objects itself. It declares one external access integration reference per product. When you allow the references in Snowsight during installation, Snowsight creates the database POSIT_TEAM_APP_DATA with the schema CONFIGURATION. It also creates one network rule per product in that schema and one external access integration per product at the account level.
Snowflake names each object after its reference and prefixes it with the application name. For an application installed as POSIT_TEAM:
| Object | Name |
|---|---|
| Workbench external access integration | POSIT_TEAM_WORKBENCH_EGRESS_EXTERNAL_ACCESS |
| Workbench network rule | POSIT_TEAM_WORKBENCH_EGRESS_NETWORK_RULE |
| Connect external access integration | POSIT_TEAM_CONNECT_EGRESS_EXTERNAL_ACCESS |
| Connect network rule | POSIT_TEAM_CONNECT_EGRESS_NETWORK_RULE |
| Package Manager external access integration | POSIT_TEAM_PACKAGE_MANAGER_EGRESS_EXTERNAL_ACCESS |
| Package Manager network rule | POSIT_TEAM_PACKAGE_MANAGER_EGRESS_NETWORK_RULE |
The fully qualified name of each network rule is POSIT_TEAM_APP_DATA.CONFIGURATION.<name>.
An application reference binds each product service to its external access integration, and each integration allows the network rule for that product. Workbench sessions and jobs use the Workbench integration. The Posit Team controller has no egress of its own. You own these objects after installation. See Egress for how to inspect and change them and for the endpoints each product needs.
Snapshots
Posit Team supports snapshots for backing up block volume data. Snapshots capture the state of a volume at a point in time and can be used to restore data after accidental deletion or corruption.