Skip to contents

Running Code Remotely

brickster provides mechanisms to run code against Databricks, below is an overview of the available of those in the package:

Method Compatible Compute Notes
SQL execution API
(db_sql_exec_*)
SQL Warehouse Lower level functions that align 1:1 with API endpoints
Command execution context manager
(db_context_manager)
Clusters
(Shared, Single User)
Higher level R6 class for command execution contexts.
Command execution API
(db_context_*)
Clusters
(Shared, Single User)
Lower level functions that align 1:1 with API endpoints
Databricks SQL Connector via reticulate
(db_sql_client)
SQL Warehouse Exposes databricks.sql connector to R.
Databricks REPL
(db_repl)
Clusters
(Shared, Single User)
Supports all notebook languages, R is only supported on single user clusters.

Databricks REPL (db_repl()) will be the focus of this article.

What is the Databricks REPL?

The REPL temporarily connects the existing R console to a Databricks cluster (via command execution APIs) and allows code in all supported languages to be sent interactively - as if it were running locally.

Getting Started

Using the REPL is simple, to start just provide cluster_id:

# start REPL
db_repl(cluster_id = "<insert cluster id>")

The REPL will check the clusters state and start the cluster if inactive. The default language is R.

After successfully connecting to the cluster you can run commands against the remote compute from the local session.

Switching Languages

The REPL has a shortcut you can enter :<language> to change the active language. You can change between the following languages:

Language Shortcut
R :r
Python :py
SQL :sql
Scala :scala
Shell :sh

When you change between languages all variables should persist unless REPL is exited.

RStudio Addins

There are three addins included with brickster for use with the REPL:

  1. Send selection to console

  2. Send current document to console

  3. Send R script to console

It’s recommended that (1) and (2) be bound to a keyboard shortcut if using the REPL frequently (‘Tools’ > ‘Modify Keyboard Shortcuts…’ > ‘Search for brickster’).

These addins behave slightly different to the standard shortcuts as they ensure that commands that span multiple lines are handled correctly via the REPL.

Limitations

  • Development environments (e.g. RStudio, Positron) won’t display variables from the remote contexts in the environment pane

  • HTML content will only render for Python, htmlwidgets rendering is restricted due to notebook limitations that require a workaround currently

  • Not designed to work with interactive serverless compute

  • Cannot persist or recover sessions