Hi all,
I'm hitting persistent Git errors in Databricks Git folders (Repos) and would appreciate any advice.
Environment
- Cloud: AWS
- Compute: serverless only
- Repo: private GitHub repository, working on a develop branch
- Contents: a Declarative Automation Bundle with Lakeflow Spark Declarative Pipelines source files (.py / .sql)
What happens
My original Git folder suddenly became unreadable. Every Git operation fails with: "This folder could not be read as a Git repository. Please try again. If the problem persists, you may need to delete and re-clone the Git folder." Running git status returns fatal: not a git repository (or any parent up to mount point /), but the .git contents look valid when I browse them (HEAD points to refs/heads/develop, and the refs, objects and config are all present). The Git dialog also shows a detached HEAD, which doesn't match the HEAD file.
I cloned the repo into a new Git folder. The clone and the checkout of develop worked, and the status was clean.
I then made a one-line edit (added a comment) to a .py file in the new folder. The next Git operation failed with: fatal: .git/index: index file open failed: Input/output error
After that, status still lists the file as modified, but the diff fails every time with: error: open("<path>/sample_trips_dbrain_lakeflow.py"): Input/output error fatal: cannot hash <path>/sample_trips_dbrain_lakeflow.py The workspace editor can still open and read the file.
Other observations
- There is no index.lock left in .git.
- After the edit, ORIG_HEAD appeared in .git and .git/config was rewritten, the same pattern I saw in the original folder before it broke.
What I've tried
- Retrying the operation and refreshing the Git dialog
- Re-cloning into a new Git folder (works until a file is edited)
Questions
- Has anyone seen "Input/output error" on .git/index or on working-tree files in Git folders? Is this a known issue?
- Could the file being part of a pipeline or bundle source folder matter (for example, the pipeline editor holding the file)?
- Is there a workaround other than repeatedly re-cloning?