8 skiLift: Connect

DBI connections, auth, and identifiers

Keywords

snowflake, R, RStudio, Posit, VS Code, workspace notebooks, skiPatrol, skiLift, mlops

8.1 Overview

skiLift is the DBI-native way to talk to Snowflake from R over the SQL REST API. Use it for dplyr pipelines, the Connections Pane, parameterized SQL, and optional Arrow/ADBC bulk I/O — without installing JDBC/ODBC or Python.

8.2 Learning Objectives

Open connections via connections.toml, JWT, PAT, or Workspace token
Understand auth modes and when to use skiLift vs sfr_connect()
Configure identifier case for interoperability with other tools

8.3 Installation

pak::pak("posit-dev/skiLift")

In Workspace, install via pre-built tarball in YAML — see Workspace Bootstrap^†. Never compile from source in Workspace.

8.4 connections.toml (recommended)

Create ~/.snowflake/connections.toml (shared with Python connector and CLI):

[default]
account   = "xy12345.us-east-1"
user      = "myuser"
authenticator = "SNOWFLAKE_JWT"
private_key_path = "~/.snowflake/rsa_key.p8"
database  = "MY_DB"
schema    = "PUBLIC"
warehouse = "MY_WH"

[dev]
account   = "xy12345.us-east-1"
user      = "myuser"
authenticator = "externalbrowser"
warehouse = "DEV_WH"
database  = "DEV_DB"
schema    = "ANALYTICS"

library(DBI)
library(skiLift)

con <- dbConnect(Snowflake())              # [default]
con <- dbConnect(Snowflake(), name = "dev")
df  <- dbGetQuery(con, "SELECT CURRENT_USER(), CURRENT_ROLE()")
dbDisconnect(con)

8.5 Programmatic Access Token (PAT)

Create a PAT in Snowsight → Admin → Security → Programmatic Access Tokens. Prefer environment variables over hard-coded tokens:

con <- dbConnect(
  Snowflake(),
  account = "xy12345.us-east-1",
  token   = Sys.getenv("SNOWFLAKE_PAT")
)

In Workspace, the managed session OAuth is preferred — reserve PATs for external CI or tools that cannot use the notebook session.

8.6 Workspace (zero config)

After setup_notebook(..., packages=["skiLift"]):

con <- dbConnect(Snowflake())  # auto-detects session token + gateway host

Bootstrap exports variables such as:

SNOWFLAKE_ACCOUNT, SNOWFLAKE_USER, SNOWFLAKE_HOST
SNOWFLAKE_ROLE, SNOWFLAKE_WAREHOUSE, SNOWFLAKE_DATABASE, SNOWFLAKE_SCHEMA (when set)

If dbConnect() fails from a %%R cell, re-run the Python bootstrap cell and confirm you are not mixing laptop connections.toml profiles with the Workspace session.

8.7 Auth modes summary

Mode	Use case
Workspace OAuth	Notebooks (zero config)
JWT key-pair	Automation, service users
PAT	Personal tokens, CI without key files
`externalbrowser`	Interactive SSO on desktop
`connections.toml`	Local dev parity with Python

8.8 Identifier case

By default skiLift uppercases unquoted table/column names in DDL/DML — matching Snowflake and ODBC behaviour:

dbWriteTable(con, "my_table", data.frame(id = 1, name = "Alice"))
dbListFields(con, "my_table")   # "ID" "NAME"

Preserve original case (e.g. lowercase for tidyverse-native pipelines):

options(skiLift.identifier_case = "preserve")

dbQuoteIdentifier() always quotes exactly what you pass; the option only affects implicit casing in table operations.

8.9 skiLift vs skiPatrol

Need	Package
ML Registry, Feature Store, doSnowflake	skiPatrol
DBI, dbplyr, Connections Pane, SQL-only	skiLift
Both in one notebook	`sfr_connect()` then `sfr_dbi_connection(conn)`

skiLift does not require Python. skiPatrol always requires Python + snowflake-ml-python for ML APIs (skiLift may still use reticulate internally for Snowpark bulk writes in Workspace — transparent to the user).

8.10 skiLift vs `odbc` (Posit / ODBC driver)

For desktop DBI + dbplyr, the other common path is CRAN odbc with Snowflake’s ODBC driver (or Posit-managed driver installs on licensed products). Both expose standard DBI; dplyr code is largely portable.

	skiLift	`odbc::snowflake()`
Transport	SQL API (+ optional ADBC)	ODBC driver
Install	R package	Driver + `odbc` package
`connections.toml`	`dbConnect(Snowflake(), name = ...)`	`dbConnect(odbc::snowflake(), connection_name = ...)`
Workspace	Default — no ODBC install	ODBC only practical if baked into CRE^†; not via bootstrap

In Workspace, ADBC (optional, often in CRE) is the optimized bulk driver path; ODBC was not viable at bootstrap time. Full comparison: RStudio, Posit & VS Code — skiLift vs ODBC.

8.11 Next steps

skiLift: Query

--- title: "skiLift: Connect" subtitle: "DBI connections, auth, and identifiers" --- ## Overview **skiLift** is the DBI-native way to talk to Snowflake from R over the [SQL REST API](../11_skilift_sql_api/index.qmd). Use it for dplyr pipelines, the Connections Pane, parameterized SQL, and optional Arrow/ADBC bulk I/O — **without** installing JDBC/ODBC or Python. ## Learning Objectives - Open connections via `connections.toml`, JWT, PAT, or Workspace token - Understand auth modes and when to use skiLift vs `sfr_connect()` - Configure identifier case for interoperability with other tools --- ## Installation {#sec-install} ```r pak::pak("posit-dev/skiLift") ``` In Workspace, install via pre-built tarball in YAML — see [Workspace Bootstrap](../06_workspace_bootstrap/index.qmd). **Never** compile from source in Workspace. --- ## connections.toml (recommended) {#sec-toml} Create `~/.snowflake/connections.toml` (shared with Python connector and CLI): ```toml [default] account = "xy12345.us-east-1" user = "myuser" authenticator = "SNOWFLAKE_JWT" private_key_path = "~/.snowflake/rsa_key.p8" database = "MY_DB" schema = "PUBLIC" warehouse = "MY_WH" [dev] account = "xy12345.us-east-1" user = "myuser" authenticator = "externalbrowser" warehouse = "DEV_WH" database = "DEV_DB" schema = "ANALYTICS" ``` ```r library(DBI) library(skiLift) con <- dbConnect(Snowflake()) # [default] con <- dbConnect(Snowflake(), name = "dev") df <- dbGetQuery(con, "SELECT CURRENT_USER(), CURRENT_ROLE()") dbDisconnect(con) ``` --- ## Programmatic Access Token (PAT) {#sec-pat} Create a PAT in **Snowsight → Admin → Security → Programmatic Access Tokens**. Prefer environment variables over hard-coded tokens: ```r con <- dbConnect( Snowflake(), account = "xy12345.us-east-1", token = Sys.getenv("SNOWFLAKE_PAT") ) ``` In **Workspace**, the managed session OAuth is preferred — reserve PATs for external CI or tools that cannot use the notebook session. --- ## Workspace (zero config) {#sec-workspace} After `setup_notebook(..., packages=["skiLift"])`: ```r con <- dbConnect(Snowflake()) # auto-detects session token + gateway host ``` Bootstrap exports variables such as: - `SNOWFLAKE_ACCOUNT`, `SNOWFLAKE_USER`, `SNOWFLAKE_HOST` - `SNOWFLAKE_ROLE`, `SNOWFLAKE_WAREHOUSE`, `SNOWFLAKE_DATABASE`, `SNOWFLAKE_SCHEMA` (when set) If `dbConnect()` fails from a `%%R` cell, re-run the Python bootstrap cell and confirm you are not mixing laptop `connections.toml` profiles with the Workspace session. --- ## Auth modes summary {#sec-auth} | Mode | Use case | |------|----------| | Workspace OAuth | Notebooks (zero config) | | JWT key-pair | Automation, service users | | PAT | Personal tokens, CI without key files | | `externalbrowser` | Interactive SSO on desktop | | `connections.toml` | Local dev parity with Python | --- ## Identifier case {#sec-case} By default skiLift **uppercases** unquoted table/column names in DDL/DML — matching Snowflake and ODBC behaviour: ```r dbWriteTable(con, "my_table", data.frame(id = 1, name = "Alice")) dbListFields(con, "my_table") # "ID" "NAME" ``` Preserve original case (e.g. lowercase for tidyverse-native pipelines): ```r options(skiLift.identifier_case = "preserve") ``` `dbQuoteIdentifier()` always quotes exactly what you pass; the option only affects implicit casing in table operations. --- ## skiLift vs skiPatrol {#sec-vs} | Need | Package | |------|---------| | ML Registry, Feature Store, doSnowflake | skiPatrol | | DBI, dbplyr, Connections Pane, SQL-only | skiLift | | Both in one notebook | `sfr_connect()` then `sfr_dbi_connection(conn)` | skiLift does **not** require Python. skiPatrol always requires Python + `snowflake-ml-python` for ML APIs (skiLift may still use reticulate internally for Snowpark bulk writes in Workspace — transparent to the user). --- ## skiLift vs `odbc` (Posit / ODBC driver) {#sec-odbc} For **desktop DBI + dbplyr**, the other common path is CRAN **`odbc`** with Snowflake's **ODBC driver** (or Posit-managed driver installs on licensed products). Both expose standard DBI; dplyr code is largely portable. | | skiLift | `odbc::snowflake()` | |--|------------|---------------------| | Transport | SQL API (+ optional ADBC) | ODBC driver | | Install | R package | Driver + `odbc` package | | `connections.toml` | `dbConnect(Snowflake(), name = ...)` | `dbConnect(odbc::snowflake(), connection_name = ...)` | | **Workspace** | Default — no ODBC install | ODBC only practical if baked into **[CRE](../09_custom_runtime_and_ml_jobs/index.qmd)**; not via bootstrap | In Workspace, **ADBC** (optional, often in CRE) is the optimized bulk driver path; ODBC was not viable at bootstrap time. Full comparison: [RStudio, Posit & VS Code — skiLift vs ODBC](../04_rstudio_posit_vscode/index.qmd#sec-odbc). --- ## Next steps [skiLift: Query](../13_skilift_query/index.qmd)