jupyter-storage-explorer-databricks-support

Databricks support for Storage Explorer, the Jupyter Storage Explorer plugin for a Databricks workspace's files: Unity Catalog volumes, workspace files and legacy DBFS, through fsspec-databricks. It also adds Databricks Connect as an engine, to read Delta Lake tables, Parquet, CSV and JSON Lines on Databricks compute.

Experimental. fsspec-databricks is experimental, and so is this plugin: it is not recommended for production use. Feedback is welcome, in this repository's issues.

Installation

pip install "jupyter-storage-explorer[databricks]"          # the plugin and its libraries
pip install "jupyter-storage-explorer[databricks-support]"  # the plugin only, with the libraries you have

The libraries are not the plugin's dependencies, so that installing it never changes those of a shared environment. The first command installs them, as does jupyter-storage-explorer-databricks-support[full]:

databricks-connect replaces pyspark: the two must not be installed together. Beside pyspark, install the plugin alone. Databricks recommends a databricks-connect of your compute's Databricks Runtime version.

The plugin is not in the explorer's all and all-supports extras: install it on purpose. It installs pyspark-support, whose actions the Databricks Connect engine shares.

Prerequisites

• fsspec-databricks reaches the workspace with the Databricks SDK's configuration: e.g. DATABRICKS_HOST and a credential, a profile of ~/.databrickscfg, or the notebook's own on Databricks. See Databricks unified authentication.
• Databricks Connect also needs compute in that configuration: a cluster, e.g. DATABRICKS_CLUSTER_ID, or serverless. See Databricks Connect.

What it does

Give the explorer fsspec-databricks' filesystem:

from fsspec_databricks import DatabricksFileSystem
from jupyter_storage_explorer import StorageExplorer

StorageExplorer(filesystems={"Databricks": DatabricksFileSystem()})

URLs work too, e.g. "dbfs://Volumes/main/sales", once fsspec_databricks.use() has made fsspec-databricks fsspec's dbfs filesystem.

The plugin gives the files locations that name the workspace, by its host, besides their path, e.g. dbfs://Volumes/main/sales/raw/a.csv:

• Two workspaces' files of the same path are told apart, also by "Paste" and the transfer policy's checks.
• Files copied in an explorer of another kernel are found again in the same workspace, even if this explorer did not show that storage, once fsspec_databricks.use() is called.

The folders of volumes, catalogs, schemas, repositories and users get icons of their own, as without the plugin. pyspark-support's code reads the files as Databricks names them, dbfs:/Volumes/....

Databricks Connect

With databricks-connect installed, the engine "Databricks Connect" reads Delta Lake tables, Parquet, CSV and JSON Lines on Databricks compute. Choose it from the status bar, as any engine.

• Files of fsspec-databricks' filesystem are read on that filesystem's workspace, with its credentials.
• Other files, e.g. on Azure or S3, are read with Databricks Connect's own configuration.
• The code it offers makes the session with DatabricksSession.builder.getOrCreate(): the workspace of your configuration, without the explorer's credentials.

Plugin options

plugin_options gives the plugin's options:

StorageExplorer(
    filesystems={"Databricks": DatabricksFileSystem()},
    plugin_options={"databricks-support": {"priority": 10}},
)

In the launcher's tabs, plugin_options() in launcher.py gives them:

def plugin_options():
    return {"databricks-support": {"enabled": False}}

Engine options

The Databricks Connect engine's options are pyspark-support's, in engine_options["databricks-connect"]:

StorageExplorer(
    filesystems={"Databricks": DatabricksFileSystem()},
    engine_options={
        "databricks-connect": {
            "*": {"session_options": {"spark.sql.ansi.enabled": "true"}},
            "csv": {"null_value": "NA"},
        },
    },
)

pyspark-support's Engine options lists them. Two differ here:

• session_options are set on the session at run time, as Databricks Connect takes no others.
• supported_protocols defaults to dbfs, s3a, abfss, abfs, wasbs, wasb and gs. The cluster cannot read the kernel's local files, so they are neither viewed nor given code with Databricks Connect.