Repository navigation
ssh: register the Databricks %pip magic on serverless environment 5+ - #7263
Merged
Merged
Conversation
The IPython startup script built PipMagicOverrides with its older constructor. On serverless environment 5 the constructor requires a logger, and on environment 6 there is no Py4J entry point, so the call failed and notebooks fell back to IPython's stock %pip. When the constructor takes a logger, build it from the driver connection the same way dbruntime's own kernel does, passing only the keyword arguments the installed runtime accepts. Older runtimes keep the existing path. Co-authored-by: Isaac <no-reply@databricks.com>
Co-authored-by: Isaac <no-reply@databricks.com>
Collaborator
✅ Integration test reportCommit: 0a842a7
|
Co-authored-by: Isaac <no-reply@databricks.com>
… keyword Only evaluate the optional constructor arguments the runtime accepts, so a missing attribute behind a newer argument doesn't break %pip registration on runtimes that don't take it. Co-authored-by: Isaac <no-reply@databricks.com>
rclarey
approved these changes
Nov 6, 2026
deco-sdk-tagging Bot
added a commit
that referenced
this pull request
Nov 9, 2026
## Release v1.25.0 ### CLI * `databricks aitools install --output json` now reports a more specific `error_category` (`MANIFEST_NOT_FOUND`, `PERMISSION_DENIED`, `FILESYSTEM`) for some failures it reported as `UNCATEGORIZED`. ([#7380](#7380)) * Suggest `databricks auth logout --profile <profile-name> --delete` when multiple profiles match the same workspace in `databricks auth docker` commands. ([#7363](#7363)) * Fix `%pip` in Jupyter notebooks opened over `databricks ssh connect` on serverless environment versions 5 and 6 and on dedicated clusters running recent Databricks Runtime versions, which fell back to the stock IPython magic instead of the Databricks one. ([#7263](#7263)) ### AI Runtime * `databricks air run` now validates the config server-side while it uploads the run artifacts, instead of before, which shortens submission. ([#7368](#7368)) * Remove `CRITICAL` as a valid `compute.priority_class` value; the AI Runtime API no longer accepts it. ([#7371](#7371)) ### Bundles * Built-in templates add the workspace ID to `workspace.host` (`?w=<id>`) when the profile or an environment variable sets it, so generated bundles select the right workspace on hosts shared by several workspaces. Welcome messages show the same URL, and `dbt-sql` also adds `?o=<id>` to the HTTP path it suggests for `dbt init`. ([#7350](#7350)) * Fix grants of a renamed or recreated catalog, schema, volume or other UC resource staying recorded under the old name, which made the next `bundle plan` propose to apply them again. ([#7345](#7345)) * Fix permissions of a recreated pipeline, app, dashboard, model serving endpoint or other resource staying recorded under the old ID, which made every later `bundle plan` propose to apply them again. ([#7361](#7361)) * Add the `postgres_data_apis` resource (Private Preview) for enabling and configuring the Data API of a Lakebase Postgres database. ([#7160](#7160)) * Fixes [#4012](#4012): registered model `aliases` are now set in Unity Catalog on deploy instead of being silently ignored. A model without an `aliases` section keeps the aliases set outside the bundle (for example by MLflow), and `aliases: []` removes them. ([#7163](#7163)) * Include variable declarations and the source of each resolved value in the `--include-locations` output of `bundle validate` and `bundle summary`. ([#7351](#7351)) * Add `.workspace.URLID` to bundle templates: the workspace ID that `.workspace.URL` adds, or empty, so templates can use it without failing when the ID is unknown. ([#7388](#7388)) * Expose the workspace of the selected profile to bundle templates as `.workspace`, with `Hostname`, `ID`, and `URL`, so templates can address a workspace on a host shared by several workspaces. A template input named `workspace` takes precedence. ([#7349](#7349)) ### Dependency Updates * Bump dependencies with known vulnerabilities. ([#7394](#7394)) * Upgrade `github.com/mattn/go-runewidth` to v0.0.31, which removes ~30ms of startup time from every command. ([#7391](#7391)) * Update the Go SDK to [v0.194.0](https://github.com/databricks/databricks-sdk-go/releases/tag/v0.194.0). ([#7371](#7371)) * Update the Go SDK to [v0.195.0](https://github.com/databricks/databricks-sdk-go/releases/tag/v0.195.0). ([#7384](#7384))
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Changes
On runtimes whose
PipMagicOverridestakes alogger(serverless environment 5+),jupyter-init.pynow builds the magic the way dbruntime's own kernel does. The driver conf, logger and environment resource client come from the driver connection, and the Py4J entry point is passed only where it exists. The constructor gained keyword arguments across runtime releases, so only the ones the installed runtime accepts are passed. This applies to any runtime with the newer constructor, including dedicated clusters on recent Databricks Runtime versions. Older runtimes keep the existing path.Why
On serverless environments 5 and 6, the IPython startup script failed to register the Databricks
%pipmagic:client.5.12):PipMagicOverrides.__init__() missing 1 required positional argument: 'logger'client.6.5):'NoneType' object has no attribute 'getDriverConf', because Py4J is off and there is no entry pointDedicated clusters on recent Databricks Runtime versions hit the same
loggererror (seen on DBR 19.14). Notebooks opened over the tunnel then fell back to IPython's stock%pip. This blocks moving the tunnel's default environment off 4.Tests
Tested on prod AWS us-west-2 CPU serverless, with this
jupyter-init.pyas the session's IPython startup script, through a real Jupyter kernel:%pipregistered byPipMagicOverrides(unchanged)PipMagicOverrides(was stock)PipMagicOverrides(was stock)%pip install humanizespark.range(3).count()On v4 a kernel started from the session can't be reached (loopback is blocked there), so v4 was checked with terminal IPython only.
Also tested on a dedicated single-user cluster on DBR 19.14 (staging AWS), through a real Jupyter kernel: before this change
%pipfell back to stock (missing 1 required positional argument: 'logger'). With it,%pipis registered byPipMagicOverrideswith an activePipMagicHandler,%pip install humanizegoes through the Databricks handler, andspark.range(3).count()returns 3.