Release v0.48.0 (#3106)

* Added `--dry-run` option for ACL migrate
([#3017](#3017)). In this
release, we have added a `--dry-run` option to the `migrate-acls`
command in the `labs.yml` file, enabling a preview of the migration
process without executing it. This feature also introduces the `hms-fed`
flag, allowing migration of HMS-FED ACLs while migrating tables. The
`ACLMigrator` class in the `application.py` file has been updated to
include new parameters, `sql_backend` and `inventory_database`, to
perform a dry run migration of Access Control Lists (ACLs).
Additionally, a new `retrieve` method has been added to the
`ACLMigrator` class to retrieve a list of grants based on the source and
destination objects, and a `CrawlerBase` class has been introduced for
fetching grants. We have also introduced a new `inferred_grants` table
in the deployment schema to store inferred grants during the migration
process.
* Added `WorkspacePathOwnership` to determine transitive owners for
files and notebooks
([#3047](#3047)). In this
release, we introduce a new class `WorkspacePathOwnership` in the
`owners.py` module to determine the transitive owners for files and
notebooks within a workspace. This class is added as a subclass of
`Ownership` and takes `AdministratorLocator` and `WorkspaceClient` as
inputs. It has methods to infer the owner from the first `CAN_MANAGE`
permission level in the access control list. We also added a new
property `workspace_path_ownership` to the existing
`HiveMetastoreContext` class, which returns a `WorkspacePathOwnership`
object initialized with an `AdministratorLocator` object and a
`workspace_client`. This addition enables the determination of owners
for files and notebooks within the workspace. The functionality is
demonstrated through new tests added to `test_owners.py`. The new tests,
`test_notebook_owner` and `test_file_owner`, create a notebook and a
workspace file and verify the owner of each using the `owner_of` method.
The `AdministratorLocator` is used to locate the administrators group
for the workspace and the `PermissionLevel` class is used to specify the
permission level for the notebook permissions.
* Added `mosaicml-streaming` to known list
([#3029](#3029)). In this
release, we have expanded the range of recognized packages in our system
by adding several new libraries to the known list in the JSON file. The
additions include `mosaicml-streaming`, `oci`, `pynacl`, `pyopenssl`,
`python-snapy`, and `zstd`. Notably, `mosaicml-streaming` has two new
entries, `simulation` and `streaming`, while the other packages have a
single entry each. This update addresses issue
[#1931](#1931) and enhances
the system's ability to identify and work with a wider variety of
packages.
* Added `msal-extensions` to known list
([#3030](#3030)). In this
release, we have added support for two new packages, `msal-extensions`
and `portalocker`, to our project. The `msal-extensions` package
includes modules for extending the Microsoft Authentication Library
(MSAL), including cache lock, libsecret, osx, persistence, token cache,
and windows. This addition enhances the library's authentication
capabilities and provides greater flexibility when working with MSAL.
The `portalocker` package offers functionalities for handling file
locking with various backends such as Redis, as well as constants,
exceptions, and utilities. This package enables developers to manage
file locking more efficiently, preventing conflicts and ensuring data
consistency. These new packages extend the range of supported packages
and functionalities for handling authentication and file locking in the
project, providing more options for software engineers to develop robust
and secure applications.
* Added `multimethod` to known list
([#3031](#3031)). In this
release, we have added support for the `multimethod` programming concept
to the library. This feature has been added to the `known.json` file,
which partially resolves issue
[#193](#193)
* Added `murmurhash` to known list
([#3032](#3032)). A new hash
function, MurmurHash, has been added to the library's supported list,
addressing part of issue
[#1931](#1931). The
MurmurHash function includes two variants, `murmurhash` and
"murmurhash.about", with distinct functionalities. The `murmurhash`
variant offers core hashing functionality, while "murmurhash.about"
contains metadata or documentation related to the MurmurHash function.
This integration enables developers to leverage MurmurHash for data
processing tasks, enhancing the library's functionality and versatility.
Users familiar with the project can now incorporate MurmurHash into
their applications and configurations, taking advantage of its unique
features and capabilities.
* Added `ninja` to known list
([#3050](#3050)). In this
release, we have added Ninja to the known list in the `known.json` file.
Ninja is a fast, lightweight build system that enables better
integration and handling within the project's larger context. This
change partially resolves issue
[#1931](#1931), which may
have been caused by challenges in integrating or using Ninja. It is
important to note that this change does not modify any existing
functionality or introduce new methods. The alteration is limited to
including Ninja in the known list, improving the management and
identification of various components within the project.
* Added `nvidia-ml-py` to known list
([#3051](#3051)). In this
release, we have added support for the `nvidia-ml-py` package to our
project. This addition consists of two components: `example` and
'pynvml'. `Example` is likely a placeholder or sample usage of the
package, while `pynvml` is a module that enables interaction with
NVIDIA's system management library (NVML) through Python. This
enhancement is a significant step towards resolving issue
[#1931](#1931), which may
require the use of NVIDIA-related tools or libraries, thereby improving
the project's functionality and capabilities.
* Added dashboard for tracking migration progress
([#3016](#3016)). This
change introduces a new dashboard for tracking migration progress in a
project, called "migration-progress", which displays real-time insights
into migration progress and facilitates planning and task division. A
new method, `_create_dashboard`, has been added to generate the
dashboard from SQL queries in a specified folder and replace database
and catalog references to match the configuration settings. The changes
include updating the install to replace the UCX catalog in queries,
adding a new object serializer, and updating integration tests and
manual testing on a staging environment. The new functionality covers
the migration of tables, views, UDFs, grants, jobs, workflow problems,
clusters, pipelines, and policies. Additionally, a new SQL file has been
added to track the percentage of various objects migrated and display
the results in the new dashboard.
* Added grant progress encoder
([#3079](#3079)). A new
`GrantsProgressEncoder` class has been introduced in the
`progress/grants.py` file to encode `Grant` objects into `History`
objects for the `migration-progress` workflow. This change includes the
addition of unit tests to ensure proper functionality and handles cases
where `Grant` objects fail to map to the Unity Catalog by adding a list
of failures to the `History` object. The commit also modifies the
`migration-progress` workflow to incorporate the new
`GrantsProgressEncoder` class, enhancing the grant processing
capabilities and improving the testing of this functionality. This
change addresses issue
[#3058](#3058), which was
related to grant progress encoding. The `GrantsProgressEncoder` class
can encode grant properties, such as the principal, action, database,
schema, table, and UDF, into a format that can be written to a backend,
ensuring successful migration of grants in the database.
* Added table progress encoder
([#3083](#3083)). In this
release, we've added a table progress encoder to the WorkflowTask
context to enhance the tracking of table-related operations in the
migration-progress workflow. This new encoder, implemented in the
TableProgressEncoder class, is connected to the sql_backend,
table_ownership, and migration_status_refresher objects. The
GrantsProgressEncoder class has been refactored to GrantProgressEncoder,
with additional parameters for improved encoding of grants. We've also
introduced the refresh_table_migration_status task to scan and record
the migration status of tables and views in the inventory, storing
results in the $inventory.migration_status inventory table. Two new unit
tests have been added to ensure proper encoding and migration status
handling. This change improves progress tracking and reporting in the
table migration process, addressing issues
[#3061](#3061) and
[#3064](#3064).
* Combine static code analysis results with historical job snapshots
([#3074](#3074)). In this
release, we have added a new method, `JobsProgressEncoder`, to the
`WorkflowTask` class in the `databricks.labs.ucx.contexts` module. This
method is used to track the progress of jobs in the context of a
workflow task, replacing the existing `jobs_progress` method which only
tracked the progress of grants. The `JobsProgressEncoder` method takes
in additional arguments, including `inventory_database`, to provide more
detailed progress tracking for jobs and is used in the `grants_progress`
method to track the progress of jobs in the context of a workflow task.
We have also added a new unit test for the `JobsProgressEncoder` class
in the `databricks.labs.ucx` project to ensure that the encoding of job
information works as expected with different types of failures and job
details. Additionally, this revision introduces the ability to include
workflow problem records in the historical job snapshots, providing
additional context for debugging and analysis. The `JobsProgressEncoder`
class is a subclass of the `ProgressEncoder` class and provides
additional functionality for tracking the progress of jobs.
* Connected `WorkspacePathOwnership` with `DirectFsAccessOwnership`
([#3049](#3049)). In this
revision, the `DirectFsAccessCrawler` class from the
`databricks.labs.ucx.source_code.directfs_access` module is imported as
`DirectFsAccessCrawler` and `DirectFsAccessOwnership`, and a new
`cached_property` called `directfs_access_ownership` is added to the
`TableCrawler` class. This property returns an instance of the
`DirectFsAccessOwnership` class, which takes in `administrator_locator`,
`workspace_path_ownership`, and `workspace_client` as arguments.
Additionally, the `DirectFsAccessOwnership` class has been updated to
determine DirectFS access ownership for a given table and connect with
`WorkspacePathOwnership`, enhancing the tool's functionality by
determining access ownership in DirectFS and improving overall system
security and permissions management. The `test_directfs_access.py` file
has also been updated to test the ownership of query and path records
using the new `DirectFsAccessOwnership` object.
* Crawlers: append snapshots to history journal, if available
([#2743](#2743)). This
commit introduces a history table to store snapshots after each crawling
operation, addressing issues
[#2572](#2572) and
[#2573](#2573). The changes
include the addition of a `HistoryLog` class, which handles appending
inventory snapshots to the history table within a specific catalog,
workspace, and run_id. The new methods also include a
`TableMigrationStatus` class with a new class variable
`__id_attributes__` to specify the attributes used to uniquely identify
a table. The `destination()` method has been added to the
`TableMigrationStatus` class to return the fully qualified name of the
destination table. Additionally, unit and integration tests have been
added and updated to ensure the functionality works as expected. The
`Table`, `Job`, `Cluster`, and `UDF` classes have been updated with a
new `history` attribute to store a string representing a problem
associated with the respective class. The `__id_attributes__` class
variable has also been added to these classes to specify the attributes
used to uniquely identify them.
* Determine ownership of tables based on grants and source code
([#3066](#3066)). In this
release, changes have been made to the `application.py` file in the
`databricks/labs/ucx/contexts` directory to improve the accuracy of
determining table ownership in the inventory. A new class
`LegacyQueryOwnership` has been added to the
`databricks.labs.ucx.framework.owners` module to determine the owner of
a table based on the queries that write to it. The `TableOwnership`
class has been updated to accept additional arguments for determining
ownership based on grants, queries, and workspace paths. The
`DirectFsAccessOwnership` class has also been updated to accept a new
`legacy_query_ownership` argument. Additionally, a new method
`owner_of_path` has been added to the `Ownership` class, and the
`LegacyQueryOwnership` class has been added as a subclass of
`Ownership`. A new file `ownership.py` has been introduced, which
defines the `TableOwnership` and `TableMigrationOwnership` classes for
determining ownership of tables and table migration records in the
inventory. These changes provide a more accurate and consistent
ownership information for tables in the inventory.
* Ensure that pipeline assessment doesn't fail if a pipeline is deleted…
([#3034](#3034)). In this
pull request, the pipelines crawler of the DLT assessment feature has
been updated to improve its resiliency in the event of a pipeline
deletion during crawling. Instead of failing, the crawler now logs a
warning and continues to crawl when a pipeline is deleted. A new test
method, `test_pipeline_disappears_during_crawl`, has been added to
verify that the crawler can handle the deletion of a pipeline after
listing the pipelines but before assessing them. The `assessment` and
`migration-progress-experimental` workflows have been modified, and new
unit tests have been added to ensure the proper functioning of the
changes. Additionally, the `test_pipeline_list_with_no_config` test case
has been added to check the behavior of the pipelines crawler when there
is no configuration present. This pull request aims to enhance the
robustness of the assessment feature and ensure its continued operation
even in the face of unexpected pipeline deletions.
* Fixed `UnicodeDecodeError` when fetching init scripts
([#3103](#3103)). In this
release, we have enhanced the error handling capabilities of the
open-source library by fixing a `UnicodeDecodeError` issue that occurred
when fetching init scripts in the `_get_init_script_data` method. To
address this, we have added `UnicodeDecodeError` and `FileNotFoundError`
to the list of exceptions handled in the method. Now, when any of these
exceptions occur, the method will return `None` and a warning message
will be logged instead of raising an unhandled exception. This change
ensures that the function operates smoothly and provides better error
handling in the library, without modifying the behavior of the
`_check_cluster_init_script` method, which remains unchanged and
continues to verify the correct setup of init scripts in the cluster.
* Fixed `UnknownHostException` on the specified KeyVault
([#3102](#3102)). In this
release, we have made significant improvements to the Azure Key Vault
integration, addressing issues
[#3102](#3102) and
[#3090](#3090). We have
resolved an `UnknownHostException` problem in a specific KeyVault and
implemented error handling for invalid Azure Key Vaults, ensuring more
robust and reliable system behavior. Additionally, we have expanded
`NotFound` exception handling to include the `InvalidState` exception.
When the Azure Key Vault is in an invalid state, the corresponding
secret will be skipped, and a warning message will be logged. This
enhancement provides a more comprehensive solution to handle various
exceptions that may arise when dealing with secrets stored in Azure Key
Vaults.
* Fixed `Unsupported schema: XXX` error on `assess_workflows`
([#3104](#3104)). The recent
change to the open-source library addresses the 'Unsupported schema:
XXX' error in the `assess_workflows` function. This was achieved by
introducing a new exception class, 'InvalidPath', in the
`WorkspaceCache` mixin, and substituting `ValueError` with `InvalidPath`
in the 'jobs.py' file. The `InvalidPath` exception is used to provide a
more specific error message for unsupported schema paths. The
`WorkspaceCache` mixin now includes an `InvalidPath` exception for
caching workspace paths. The error handling in the 'jobs.py' file has
been modified to raise `InvalidPath` instead of `ValueError` for better
error messages. Additionally, the 'test_cached_workspace_path.py' file
has updates for testing the `WorkspaceCache` object, including the
addition of the `InvalidPath` exception for non-absolute paths, and a
new test function for this exception. The `WorkspaceCache` class has an
ellipsis in the `__init__` method, indicating additional initialization
code not shown in this diff.
* Fixed `assert curr.location is not None`
([#3105](#3105)). In this
release, we have addressed a potential issue in the
`_external_locations` method which failed to check if the location of
the current Hive table is `None` before proceeding. This oversight could
result in unnecessary exceptions when accessing the location of a Hive
table. To rectify this, we have introduced a check for `None` that will
bypass the current iteration of the loop if the location is not set,
thereby improving the robustness of the code. The method continues to
return a list of `ExternalLocation` objects, each representing a Hive
table or partition location with the corresponding number of tables or
partitions present. The `ExternalLocation` class remains unchanged in
this commit. This improvement will ensure that the method functions
smoothly and avoids errors when dealing with Hive tables that do not
have a location set.
* Fixed dynamic import issue
([#3053](#3053)). In this
release, we've addressed an issue related to dynamic import inference in
our open-source library. Previously, the code did not infer import names
when using `importlib.import_module(some_name)`. This has been resolved
by implementing a new method, `_make_sources_for_import_call_node`,
which infers the import name from the provided node argument.
Additionally, we've introduced new functions, `get_global(self, name:
str)`, `_adjust_node_for_import_member(self, name: str, match_node:
type, node: NodeNG)`, and updated the `_matches(self, node: NodeNG,
depth: int)` method to handle attributes as global names. A new unit
test, `test_graph_imports_dynamic_import()`, has been added to ensure
the proper functioning of the dynamic import feature. Moreover, a new
function `is_from_module` has been introduced to check if a given name
is from a specific module. This commit, co-authored by Eric Vergnaud,
significantly enhances the code's ability to infer imports in dynamic
import scenarios.
* Fixed issue with migrating `MANAGED` hive_metastore table to UC for
`CONVERT_TO_EXTERNAL` scenario
([#3020](#3020)). This
change updates the process for converting a managed Hive Metastore (HMS)
table to external in the CONVERT_TO_EXTERNAL scenario. The functionality
is split into a separate workflow task, executed from a non-Unity
Catalog (UC) cluster, and is tested with unit and integration tests. The
migrate table function for external sync ensures the table is migrated
as external to UC post-conversion. The changes include adding a new
workflow and modifying an existing one, and updates the existing
workflow to rename the migrate_tables function to
convert_managed_hms_to_external. The new function handles the conversion
of managed HMS tables to external, and updates the object_type property
of the table in the inventory database to `EXTERNAL` after the
conversion is completed. The pull request resolves issue
[#2840](#2840) and removes
the existing functionality of applying grants during the migration
process.
* Fixed issue with table location on storage root
([#3094](#3094)). In this
release, we have implemented changes to address an issue related to the
incorrect identification of the parent folder as an external location
when there is a single table with a prefix that matches a parent folder.
Additionally, we have improved the storage and retrieval of table
locations in the root directory of a storage service by adding support
for additional S3 bucket URL formats in the unit tests for the Hive
Metastore. This includes handling S3 bucket URLs that do not include a
specific file or path, and those with a path that does not include a
file. We have also added new test cases for these URL formats and
modified existing ones to include them. These changes ensure correct
identification of external locations and improve functionality and
flexibility of the Hive Metastore's support for external table
locations. The new methods added are not explicitly stated, but they
likely involve functions for parsing and processing the new S3 bucket
URL formats.
* Fixed snapshot loading for DFSA and used-table crawlers
([#3046](#3046)). This
commit resolves issues related to snapshot loading for the DFSA and
used-table crawlers when using the spark-based lsql backend. The root
cause was the use of `.as_dict()` to convert rows to dictionaries, which
is unavailable in the spark-based lsql backend. The fix involves
replacing this method with `.asDict()`. Additionally, integration and
unit tests were updated to include snapshot loading for these crawlers,
and a typo in a test name was corrected. The changes are confined to the
test_queries.py file and do not affect other parts of the project. No
new methods were added, and existing functionality changes were limited
to updating the snapshot loading process.
* Ignore failed inference codes when presenting results to Databricks
Runtime ([#3087](#3087)). In
this release, the `lsp_plugin.py` file has been updated in the
`databricks/labs/ucx/source_code` directory to improve the user
experience in the notebook editor. The changes include disabling certain
advice codes from being propagated, specifically:
'cannot-autofix-table-reference', 'default-format-changed-in-dbr8',
'dependency-not-found', 'not-supported',
'notebook-run-cannot-compute-value', 'sql-parse-error',
'sys-path-cannot-compute-value', and 'unsupported-magic-line'. A new
variable `DEBUG_MESSAGE_CODES` has been introduced to store the list of
advice codes to be ignored, and the list comprehension that creates
`diagnostics` in the `pylsp_lint` function has been updated to exclude
these codes. These updates aim to reduce the number of unnecessary error
messages and improve the accuracy of the linter for supported codes.
* Improve scan tables in mounts
([#2767](#2767)). In this
release, the `scan-tables-in-mounts` functionality in the hive metastore
has been significantly improved, providing a more robust and
comprehensive solution. Previously, the implementation skipped most
directories, only finding 8 tables, but this issue has been addressed,
allowing the updated version to parse many more tables. The commit
includes bug fixes and the addition of new unit tests. The reviewer is
encouraged to refactor the code in future iterations to use the `os`
module instead of `dbutils` for listing directories, enabling
parallelization and improving scalability. The commit resolves issue
[#2540](#2540) and updates
the `scan-tables-in-mounts-experimental` workflow. While manual and unit
tests have been added and verified, integration tests are still pending
implementation. The co-author of this commit is Dan Zafar.
* Removed `WorkflowLinter` as it is part of the `Assessment` workflow
([#3036](#3036)). In this
release, the `WorkflowLinter` has been removed as it is now integrated
into the `Assessment` workflow, addressing issue
[#3035](#3035). This change
simplifies the codebase, removing the need for a separate linter while
maintaining essential functionality for ensuring Unity Catalog
compatibility. The linter's functionality has been merged with other
parts of the assessment workflow, with results persisted in the
`$inventory_database.workflow_problems` and
`$inventory_database.directfs_in_paths` tables. The `assess_workflows`
and `assess_dashboards` methods have been updated accordingly, removing
`WorkflowLinter` usage. Additionally, the `ExperimentalWorkflowLinter`
class has been removed from the `workflows.py` file, along with its
associated methods `lint_all_workflows` and `lint_all_queries`. The
`test_running_real_workflow_linter_job` function has also been removed
due to the integration of the `WorkflowLinter` into the `Assessment`
workflow. Manual testing has been conducted to ensure the correctness of
these changes and the continued proper functioning of the assessment
workflow.
* Updated permissions crawling so that it doesn't fail if a secret scope
disappears during crawling
([#3070](#3070)). This
commit enhances the open-source library by updating the permissions
crawling process for secret scopes, addressing the issue of task failure
when a secret scope disappears before ACL retrieval. The `assessment`
workflow has been modified to incorporate these updates, and new unit
tests have been added, including one that simulates the disappearance of
a secret scope during crawling. The `PermissionsCrawler` class and the
`Threads.gather` method have been improved to handle such cases, logging
a warning instead of failing the task. The return type of the
`get_crawler_tasks` method has been updated to Iterable[Callable[[],
Permissions | None]]. These changes improve the reliability and
robustness of the permissions crawling process for secret scopes,
ensuring task completion in the face of unexpected scope disappearances.
* Updated sqlglot requirement from <25.26,>=25.5.0 to >=25.5.0,<25.27
([#3041](#3041)). In this
pull request, we have updated the sqlglot library requirement to
incorporate the latest version, which includes various bug fixes,
refactors, and exciting new features. The latest version now supports
the TO_DOUBLE and TRY_TO_TIMESTAMP functions in Snowflake and the
EDIT_DISTANCE (Levinshtein) function in BigQuery. Moreover, we've
addressed an issue with the ARRAY JOIN function in Clickhouse and made
changes to the hive dialect hierarchy. We encourage users to update to
this latest version to benefit from these enhancements and fixes,
ensuring optimal performance and functionality of the library.
* Updated sqlglot requirement from <25.27,>=25.5.0 to >=25.5.0,<25.28
([#3048](#3048)). In this
release, we have updated the requirement for the `sqlglot` library to a
version greater than or equal to 25.5.0 and less than 25.28. This change
was made to allow for the use of the latest features and bug fixes
available in 'sqlglot', while avoiding the breaking changes that were
introduced in version 25.27. The new version of `sqlglot` offers several
improvements, including but not limited to enhanced query optimization,
expanded support for various SQL dialects, and better error handling. We
recommend that all users upgrade to the latest version of `sqlglot` to
take advantage of these new features and improvements.
* Updated sqlglot requirement from <25.28,>=25.5.0 to >=25.5.0,<25.29
([#3093](#3093)). This
release includes an update to the `sqlglot` dependency, changing the
version requirement from 25.5.0 up to but excluding 25.28, to a range
that includes 25.5.0 up to but excluding 25.29. This change allows for
the use of the latest `sqlglot` version and includes all the updates and
bug fixes from this library since the previous version. The pull request
provides a list of changes made in `sqlglot` since the previous version,
as well as a list of relevant commits. Dependabot has been configured to
handle any merge conflicts for this pull request and includes commands
to trigger various Dependabot actions. This update was made by
Dependabot and is indicated by a signed-off-by line.

Dependency updates:

* Updated sqlglot requirement from <25.26,>=25.5.0 to >=25.5.0,<25.27
([#3041](#3041)).
* Updated sqlglot requirement from <25.27,>=25.5.0 to >=25.5.0,<25.28
([#3048](#3048)).
* Updated sqlglot requirement from <25.28,>=25.5.0 to >=25.5.0,<25.29
([#3093](#3093)).