## Changes
This PR adds support for UC volumes to DABs.
### Can I use a UC volume managed by DABs in `artifact_path`?
Yes, but we require the volume to exist before being referenced in
`artifact_path`. Otherwise you'll see an error that the volume does not
exist. For this case, this PR also adds a warning if we detect that the
UC volume is defined in the DAB itself, which informs the user to deploy
the UC volume in a separate deployment first before using it in
`artifact_path`.
We cannot create the UC volume and then upload the artifacts to it in
the same `bundle deploy` because `bundle deploy` always uploads the
artifacts to `artifact_path` before materializing any resources defined
in the bundle. Supporting this in a single deployment requires us to
migrate away from our dependency on the Databricks Terraform provider to
manage the CRUD lifecycle of DABs resources.
### Why do we not support `preset.name_prefix` for UC volumes?
UC volumes will not have a `dev_shreyas_goenka` prefix added in `mode:
development`. Configuring `presets.name_prefix` will be a no-op for UC
volumes. We have decided not to support prefixing for UC resources. This
is because:
1. UC provides its own namespace hierarchy that is independent of DABs.
2. Users can always manually use `${workspace.current_user.short_name}`
to configure the prefixes manually.
Customers often manually set up a UC hierarchy for dev and prod,
including a schema or catalog per developer. Thus, it's often
unnecessary for us to add prefixing in `mode: development` by default
for UC resources.
In retrospect, supporting prefixing for UC schemas and registered models
was a mistake and will be removed in a future release of DABs.
## Tests
Unit, integration test, and manually.
### Manual Testing cases:
1. UC volume does not exist:
```
➜ bundle-playground git:(master) ✗ cli bundle deploy
Error: failed to fetch metadata for the UC volume /Volumes/main/caps/my_volume that is configured in the artifact_path: Not Found
```
2. UC Volume does not exist, but is defined in the DAB
```
➜ bundle-playground git:(master) ✗ cli bundle deploy
Error: failed to fetch metadata for the UC volume /Volumes/main/caps/managed_by_dab that is configured in the artifact_path: Not Found
Warning: You might be using a UC volume in your artifact_path that is managed by this bundle but which has not been deployed yet. Please deploy the UC volume in a separate bundle deploy before using it in the artifact_path.
at resources.volumes.bar
in databricks.yml:24:7
```
---------
Co-authored-by: Pieter Noordhuis <pieter.noordhuis@databricks.com>
## Changes
Prior to this change, the output directory was part of the `renderer`
type and passed down to every `file` it produced. Every file knew its
absolute destination path. This is incompatible with the use of a filer,
where all operations are automatically anchored to some base path.
To make this compatible, this change updates:
* the `file` type to only know its own path relative to the instantiation root,
* the `renderer` type to no longer require or pass along the output directory,
* the `persistToDisk` function to take a context and filer argument,
* the `filer.WriteMode` to represent permission bits
## Tests
* Existing tests pass.
* Manually confirmed template initialization works as expected.
## Changes
The workspace extensions filer should not read or stat a notebook called
`foo` if the user calls `.Stat(ctx, "foo")`.
Instead, the filer should return a file not found error. This is because
the contract for the workspace extensions filer is to only work for
notebooks when the file path / name includes the extension (example:
`foo.ipynb` or `foo.sql` instead of just `foo`)
## Tests
Integration tests.
## Changes
We had a number of copies of test helpers for `io/fs` in the repository.
This change consolidates all of them to use the `libs/fakefs` package.
## Tests
Unit tests pass.
## Changes
### Background
The workspace import APIs recently added support for importing Jupyter
notebooks written in R, Scala, or SQL, that is non-Python notebooks.
This now works for the `/import-file` API which we leverage in the CLI.
Note: We do not need any changes in `databricks sync`. It works out of
the box because any state mapping of local names to remote names that we
store is only scoped to the notebook extension (i.e., `.ipynb` in this
case) and is agnostic of the notebook's specific language.
### Problem this PR addresses
The extension-aware filer previously did not function because it checks
that a `.ipynb` notebook is written in Python. This PR relaxes that
constraint and adds integration tests for both the normal workspace
filer and extensions aware filer writing and reading non-Python `.ipynb`
notebooks.
This implies that after this PR DABs in the workspace / CLI from DBR
will work for non-Python notebooks as well. non-Python notebooks for
DABs deployment from local machines already works after the platform
side changes to the API landed, this PR just adds integration tests for
that bit of functionality.
Note: Any platform side changes we needed for the import API have
already been rolled out to production.
### Before
DABs deploy would work fine for non-Python notebooks. But DABs
deployments from DBR would not.
### After
DABs deploys both from local machines and DBR will work fine.
## Testing
For creating the `.ipynb` notebook fixtures used in the integration
tests I created them directly from the VSCode UI. This ensures high
fidelity with how users will create their non-Python notebooks locally.
For Python notebooks this is supported out of the box by VSCode but for
R and Scala notebooks this requires installing the Jupyter kernel for R
and Scala on my local machine and using that from VSCode.
For SQL, I ended up directly modifying the `language_info` field in the
Jupyter metadata to create the test fixture.
### Discussion: Issues with configuring language at the cell level
The language metadata for a Jupyter notebook is standardized at the
notebook level (in the `language_info` field). Unfortunately, it's not
standardized at the cell level. Thus, for example, if a user changes the
language for their cell in VSCode (which is supported by the standard
Jupyter VSCode integration), it'll cause a runtime error when the user
actually attempts to run the notebook. This is because the cell-level
metadata is encoded in a format specific to VSCode:
```
cells: []{
"vscode": {
"languageId": "sql"
}
}
```
Supporting cell level languages is thus out of scope for this PR and can
be revisited along with the workspace files team if there's strong
customer interest.
## Changes
This adds diagnostics for collaborative (production) deployment
scenarios, including:
- Bob deploys a bundle that is normally deployed by Alice, but this
fails because Bob can't write to `/Users/Alice/.bundle`.
- Charlie deploys a bundle that is normally deployed by Alice, but this
fails because he can't create a new pipeline where Alice would be the
owner.
- Alice deploys a bundle where she didn't list herself as one of the
CAN_MANAGE users in permissions. That can work, but is probably a
mistake.
## Tests
Unit tests, manual testing.
## Changes
`TestAccFilerWorkspaceFilesExtensionsErrorsOnDupName` recently started
failing in our nightlies because the upstream `import` API was changed
to [prohibit conflicting file
paths](https://docs.databricks.com/en/release-notes/product/2024/august.html#files-can-no-longer-have-identical-names-in-workspace-folders).
Because existing conflicting file paths can still be grandfathered in,
we need to retain coverage for the test. To do this, this PR:
1. Removes the failing
`TestAccFilerWorkspaceFilesExtensionsErrorsOnDupName`
2. Add an equivalent unit test with the `list` and `get-status` API
calls mocked.
## Changes
This PR adds autocomplete for cat, cp, ls, mkdir and rm.
The new completer can do completion for any `Filer`. The command
completion for the `sync` command can be moved to use this general
completer as a follow-up.
## Tests
- Tested manually against a workspace
- Unit tests
## Changes
By default, construct a read/write instance. If constructed in read-only
mode, the underlying filer is wrapped in a readahead cache.
## Tests
* Filer integration tests pass.
* Manual test that caching is enabled when running on WSFS.
## Changes
The reason this readahead cache exists is that we frequently need to
recursively find all files in the bundle root directory, especially for
sync include and exclude processing. By caching the response for every
file/directory and frontloading the latency cost of these calls, we
significantly improve performance and eliminate redundant operations.
## Tests
* [ ] Working on unit tests
## Changes
This PR fixes a performance bug that led downloaded files (e.g. with
`databricks fs cp dbfs:/Volumes/.../somefile .`) to be buffered in
memory before being written.
Results from profiling the download of a ~100MB file:
Before:
```
Type: alloc_space
Showing nodes accounting for 374.02MB, 98.50% of 379.74MB total
```
After:
```
Type: alloc_space
Showing nodes accounting for 3748.67kB, 100% of 3748.67kB total
```
Note that this fix is temporary. A longer term solution should be to use
the API provided by the Go SDK rather than making an HTTP request
directly from the CLI.
fix#1575
## Tests
Verified that the CLI properly download the file when doing the
profiling.
## Changes
This consolidates the two separate status calls into one.
The extension-aware filer now doesn't need the direct API client anymore
and fully relies on the underlying filer.
## Tests
* Unit tests.
* Ran the filer integration tests manually.
## Changes
Not doing this meant file system traversal ended upon reaching a Git
folder. By marking these objects as a directory globbing traverses into
these folders as well.
## Tests
Added a unit test for coverage.
## Changes
This worked fine if the notebooks are located in the filer's root and
didn't if they are nested in a directory.
This change adds test coverage and fixes the underlying issue.
## Tests
Ran integration test manually.
## Changes
From the [documentation](https://pkg.go.dev/os#IsNotExist) on the
functions in the `os` package:
> This function predates errors.Is. It only supports errors returned by
the os package.
> New code should use errors.Is(err, fs.ErrNotExist).
This issue surfaced while working on using a different `vfs.Path`
implementation that uses errors from the `fs` package. Calls to
`os.IsNotExist` didn't return true for errors that wrap
`fs.ErrNotExist`.
## Tests
n/a
## Changes
This PR adds a filer that'll allow us to read notebooks from the WSFS
using their full paths (with the extension included). The filer relies
on the existing workspace filer (and consequently the workspace
import/export/list APIs).
Using this filer along with a virtual filesystem layer
(https://github.com/databricks/cli/pull/1452/files) will allow us to use
our custom implementation (which preserves the notebook extensions)
rather than the default mount available via DBR when the CLI is run from
DBR.
## Tests
Integration tests.
---------
Co-authored-by: Pieter Noordhuis <pieter.noordhuis@databricks.com>
## Changes
The prior join call calls `filepath.Join` which returns a cleaned
result.
Path cleaning, in turn, calls `filepath.FromSlash`.
## Tests
* Unit tests.
## Changes
```
shreyas.goenka@THW32HFW6T cli % databricks fs -h
Commands to do file system operations on DBFS and UC Volumes.
Usage:
databricks fs [command]
Available Commands:
cat Show file content.
cp Copy files and directories.
ls Lists files.
mkdir Make directories.
rm Remove files and directories.
```
This PR adds support for UC Volumes to the fs commands. The fs commands
for UC volumes work the same as they currently do for DBFS. This is
ensured by running the same test matrix we across both DBFS and UC
Volumes versions of the fs commands.
## Tests
Support for UC volumes is tested by running the same tests as we did
originally for DBFS commands. The tests require a `main` catalog to
exist in the workspace, which does in our test workspaces environments
which have the `TEST_METASTORE_ID` environment variable set.
For the Files API filer, we do the same by running mostly common tests
to ensure the filers for "local", "wsfs", "dbfs" and "files API" are
consistent.
The tests are also made to all run in parallel to reduce the time taken.
To ensure the separation of the tests, each test creates its own UC
schema (for UC volumes tests) or DBFS directories (for DBFS tests).
## Changes
A bug in the code that pulls the remote state could cause the local
state to be empty instead of a copy of the remote state. This happened
only if the local state was present and stale when compared to the
remote version.
We correctly checked for the state serial to see if the local state had
to be replaced but didn't seek back on the remote state before writing
it out. Because the staleness check would read the remote state in full,
copying from the same reader would immediately yield an EOF.
## Tests
* Unit tests for state pull and push mutators that rely on a mocked
filer.
* An integration test that deploys the same bundle from multiple paths,
triggering the staleness logic.
Both failed prior to the fix and now pass.
## Changes
* Update Go SDK to v0.19.0
* Update commands per OpenAPI spec from Go SDK
* Incorporate `client.Do()` signature change to include a (nil) header
map
* Update `workspace.WorkspaceService` mock with permissions methods
* Skip `files` service in codegen; already implemented under the `fs`
command
## Tests
Unit and integration tests pass.
## Changes
Use the download method from the SDK in the read method for the WSFS
implementation of the filer interface.
Closes#452.
## Tests
Tested by existing integration tests
## Changes
This PR removes the stat call and instead relies on errors returned by
the go SDK to return the appropriate errors
## Tests
Tested using existing filer integration tests
## Tests
New integration test for the read/write parts of the other filers. The
integration test cannot be shared just yet because the Files API doesn't
include support for creating/listing/removing directories yet.
## Changes
Local file reads on Windows require the file handle to be closed after
using it. This commit includes an interface change to return an
`io.ReadCloser` from `Read` to accommodate this.
## Tests
The existing integration tests for the filer interface all pass.
## Changes
This PR:
1. Adds the export-dir command
2. Changes filer.Read to return an error if a user tries to read a
directory
3. Adds returning internal file structures from filer.Stat().Sys()
## Tests
Integration tests and manually
## Changes
This captures the recursive deletion of a directory tree in the filer interface.
Prompted by #433.
## Tests
Integration tests pass (ran the filer ones on AWS and Azure).
## Changes
This enables the use of `io/fs` functions `fs.Glob` and `fs.WalkDir`
with filers.
We can't use `fs.FS` as the standard interface instead of `filer.Filer` because:
1. It was made for reading from filesystems only, not writing
2. It doesn't take a context for the core functions
Therefore a wrapper will do.
## Tests
* Added unit tests to cover the adapter through a fake filer.
* Manually ran `fs.WalkDir` against both WSFS and DBFS filers.
## Changes
The pattern `errors.Is(err, fs.ErrNotExist)` is common to check for an
error type.
Errors can implement `Is(error) bool` with a custom equivalence checker.
## Tests
New asserts all pass in the integration test.
Adds a DBFS implementation of the `filer.Filer` interface.
The integration tests are reused between the workspace filesystem and
DBFS implementations to ensure identical behavior.
Summary:
* All remote path arguments for deployer and locker are now relative to
root specified at initialization
* The workspace client is now a struct field so it doesn't have to be
passed around
shreyas-goenka
Pieter Noordhuis
Lennart Kats (databricks)
andersrexdb
Renaud Hartert
Andrew Nester