* Add fixtures for setting and resettign flags for unit tests
* Remove unnecessary `set_from_args` in non `unittest.TestCase` based unit tests
In the previous commit we added a pytest fixture which sets and tears down
the global flags arg via `set_from_args` for every pytest based unit test.
Previously we had added a `set_from_args` in tests or test files to reset
the global flags from if they were modified by a previous test. This is no
longer necessary because of the work done in the previous commit.
Note: We did not modify any tests that use the `unittest.TestCase` class
because they don't use pytest fixtures. Thus those tests need to continue
operating as they currently do until we shift them to pytest unit tests.
* Utilize the new `args_for_flags` fixture for setting of flags in `test_contracts_graph_parsed.py`
* Convert `test_compilation.py` from `TestCase` tests to pytest tests
We did this so in the next commit we can drop the unnecessary `set_from_args`
in the next commit. That will be it's own commit because converting these
tests is a restructuring that doing separately makes things easier to follow.
That is to say, all changes in this commit were just to convert the tests to
pytest, no other changes were made.
* Drop unnecessary `set_from_args` in `test_compilation.py`
* Add return types to all methods in `test_compilation.py`
* Reduce imports from `compilation` in `test_compilation.py`
* Update `test_logging.py` now that we don't need to worry about global flags
* Conditionally import `Generator` type for python 3.8
In python 3.9 `Generator` was moved to `collections.abc` and deprecated
in `typing`. We still support 3.8 and thus need to be conditionally
importing `Generator`. We should remove this in the future when we drop
support for 3.8.
* Add more accurate RSS high water mark measurement for Linux
* Add changelog entry.
* Checks to avoid exception based flow control, per review.
* Fix review nit.
* Add unit test to assert `setup_config_logger` clears the event manager state
* Move `setup_event_logger` tests from `test_functions.py` to `test_logging.py`
* Move `EventCatcher` to `tests.utils` for use in unit and functional tests
* Update fixture mocking global event manager to instead clear it
Previously we had started _mocking_ the global event manager. We did this
because we though that meant anything we did to the global event manager,
including modifying it via things like `setup_event_logger`, would be
automatically cleaned up at the end of any test using the fixture because
the mock would go away. However, this understanding of fixtures and mocking
was incorrect, and the global event manager wouldn't be properly isolated/reset.
Thus we changed to fixture to instead cleanup the global event manager before
any test that uses it and by using `autouse=True` in the fixture definition
we made it so that every unit test uses the fixture.
Note this will no longer be viabled if we every multi-thread our unit testing as
the event manager isn't actually isolated, and thus two tests could both modify
the event manager at the same time.
* Add test for different `write_perf_info` values to `get_full_manifest`
* Add test for different `reset` values to `get_full_manifest`
* Abstract required mocks for `get_full_manifest` tests to reduce duplication
There are a set of required mocks that `get_full_manifest` unit tests need.
Instead of doing these mocks in each test, we've abstracted these mocks into
a reusable function. I did try to do this as a fixture, but for some reaosn
the mocks didn't actually propagate when I did that.
* Add test for different `PARTIAL_PARSE_FILE_DIFF` values to `get_full_manifest`
* Refactor mock fixtures in `test_manifest.py` to make them more widely available
* Convert `set_required_mocks` of `TestGetFullManifest` into a fixture
This wasn't working before, but it does now. Not sure why.
This was done by running `pre-commit run --all`. That this was needed
is a temporary glitch in how our `Tests and Code Checks` github action
works on PRs. Basically we added `isort` to the pre-commit hooks recently, and
this does additional linting/formatting on our imports.
People reasonably have branches which were started prior to `isort` being
part of the pre-commit hooks on main. Thus, unless those branches get caught
up to main, the github action on associated PRs won't run `isort` because
it doesn't exist on those branchs. Once everyone gets their local `main`
branch updated (I suspect this might take a few days) this problem will go
away.
* Add `isort` as a dev-req and pre-commit hook
The tool `isort` reorders imports to be in alphabetical order. I've
added it because our imports in most files are in random order. The lack
of order meant that:
- sometimes the same module would be imported from twice
- figuring out if a a module was already being imported from took
longer
In the next commit I'll actually run isort to order everything. The best
part is that when developing, we don't have to put them in correct order.
Though you can if you want. However, `isort` will take care of re-ordering
things at commit time :)
* Improve isort functionality by setting initial `.isort.cfg`
Specifically we set two config values: `extend_skip_glob` and `known_first_party`.
The `extend_skip_glob` extends the default skipped paths. The defaults can be seen
here https://pycqa.github.io/isort/docs/configuration/options.html#skip. We are skipping
third party stubs because these are more so provided (I believe). We are skipping
`.github` and `scripts` as they feel out of scope and things we can be less strict with.
The `known_first_party` setting makes it so that these imports get grouped separately from
all other imports, which is useful visually of "this comes from us" vs "this comes from
someone/somewhere else".
* Add profile `black` to isort config
I was seeing some odd behavior where running pre-commit, adding the modified
files, and then running pre-commit again would result making more modifications
to some of the same files. This felt odd. You shouldn't have to run pre-commit
more multiple times for it to eventually come to a final "solution". I believe
the problem was because we are using the tool `black` to format things, but weren't
registering the black profile with `isort` this lead to some conflicting formatting
rules, and the two tools had to negotiate a few times before being both satisfied.
Registering the profile `black` with `isort` resolved this problem.
* Reorder, merge-duplicate, and format module imports using `isort`
This was done by running `pre-commit run --all`. I ran it separately from
the commit process itself because I wanted to run it against all files
instead of only changed files.
Of note, this not only reordered and formatted our imports. But we also
had 60 distinct duplicate module import paths across 50 files, which this
took care of. When I say "distinct duplicate module import paths" I mean
when `from x.y.z import` was imported more than once in a single file.
* add support for explicit nulls for loaded_at_field
* add test
* changelog
* add parsing for tests
* centralize logic a bit
* account for sources being None
* fix bug
* remove new field from SourceDefinition
* add validation for empty string, mroe tests
* Move deferral from task to manifest loading + RefResolver
* dbt clone must specify --defer
* Fix deferral for unit test type deteection
* Add changelog
* Move merge_from_artifact from end of parsing back to task before_run to reduce scope of refactor
* PR review. DeferRelation conforms to RelationConfig protocol
* Add test case for #10017
* Update manifest v12 in test_previous_version_state
---------
Co-authored-by: Michelle Ark <michelle.ark@dbtlabs.com>
* Change agate upper bound to v1.10
* Add changelog.
* update lower pin
* for testing
* put back dev requirement
* move the lower pin back to 1.7
---------
Co-authored-by: Emily Rockman <emily.rockman@dbtlabs.com>
* Refactor test class `EventCatcher` into utils to make accessible to other tests
* Raise minimum version of dbt_common to 1.0.2
We're going to start depending on `silence` existing as a attr of
`WarnErrorOptions`. The `silence` attr was only added in dbt_common
1.0.2, thus that is our new minimum.
* Add ability to silence warnings from `warn_error_options` CLI arguments
* Add `flush` to `EventCatcher` test util, and use in `test_warn_error_options`
* Add tests to `TestWarnErrorOptionsFromCLI` for `include` and `exclude`
* Test support for setting `silence` of `warn_error_options` in `dbt_project` flags
Support for `silence` was _automatically_ added when we upgraded to dbt_common 1.0.2.
This is because we build the project flags in a `.from_dict` style, which is cool. In
this case it was _automatically_ handled in `read_project_flags` in `project.py`. More
specifically here bcbde3ac42/core/dbt/config/project.py (L839)
* Add tests to `TestWarnErrorOptionsFromProject` for `include` and `exclude`
Typically we can't have multiple tests in the same `test class` if they're
utilizing/modifying file system fixtures. That is because the file system
fixtures are scoped to test classes, so they don't reset inbetween tests within
the same test class. This problem _was_ affectin these tests as they modify the
`dbt_project.yml` file which is set by a class based fixture. To get around this,
because I find it annoying to create multiple test classes when the tests really
should be grouped, I created a "function" scoped fixture to reset the `dbt_project.yml`.
* Update `warn_error_options` in CLI args to support `error` and `warn` options
Setting `error` is the same as setting `include`, but only one can be specified.
Setting `warn` is the same as setting `exclude`, but only one can be specified.
* Update `warn_error_options` in Project flags to support `error` and `warn` options
As part of this I refactored `exclusive_primary_alt_value_setting` into an upstream
location `/config/utils`. That is because importing it in `/config/project.py` from
`cli/option_types.py` caused some circular dependency issues.
* Use `DbtExclusivePropertyUseError` in `exclusive_primary_alt_value_setting` instead of `DbtConfigError`
Using `DbtConfigError` seemed reasonable. However in order to make sure the error
got raised in `read_project_flags` we had to mark it to be `DbtConfigError`s to be
re-raised. This had the uninteded consequence of reraising a smattering of errors
which were previously being swallowed. I'd argue that if those are errors we're
swallowing, the functions that raise those errors should be modified perhaps to
conditionally not raise them, but that's not the world we live in and is out of
scope for this branch of work. Thus instead we've created a error specific to
`WarnErrorOptions` issues, and now use, and catch for re-raising.
* Add unit tests for `exclusive_primary_alt_value_setting` method
I debated about parametrizing these tests, and it can be done. However,
I found that the resulting code ended up being about the same number of
lines and slightly less readable (in my opinion). Given the simplicity of
these tests, I think not parametrizing them is okay.
Letting the dbt version be dynamic in the project fixture previously was
causing some tests to break whenever the version of dbt actually got updated,
which isn't great. It'd be super annoying to have to always update tests
affected by this. To get around this we've gone and hard coded the dbt version
in the profile. The alternative was to interpolate the version during comparison
during the relevant tests, which felt less appealing.
* Move `tests/unit/test_yaml_renderer.py` to `tests/unit/parser/test_schema_renderer.py`
* Move `tests/unit/test_unit_test_parser.py` to `tests/unit/parser/test_unit_tests.py`
* Convert `tests/unit/test_tracking.py` to use pytest fixtures
* Delete `tests/unit/test_sql_result.py` as it was moved to `dbt-adapters`
* Move `tests/unit/test_semantic_models.py` to `tests/unit/graph/test_nodes.py
* Group tests of `SemanticModel` in `test_nodes.py` into a `TestSemanticModel` class
* Move `tests/unit/test_selector_errors.py` to `tests/unit/config/test_selectors.py`
* Add `Project` fixture for unit tests and test `Project` class methods
* Move `Project.__eq__` unit tests to new pytest class testing
* Move `Project.hashed_name` unit test to pytest testing class
* Rename some testing class in `test_project.py` to align with testing split
* Refactor `project` fixture to make accessible to other unit tests
* simplify dockerfile, eliminate references to adapter repos as they will be handled in those repos
* keep dbt-postgres target for historical releases of dbt-postgres
* update third party image to pip install conditionally
* Add event type for deprecation of spaces in model names
* Begin emitting deprecation warning for spaces in model names
* Only warn on first model name with spaces unless `--debug` is specified
For projects with a lot of models that have spaces in their names, the
warning about this deprecation would be incredibly annoying. Now we instead
only log the first model name issue and then a count of how many models
have the issue, unless `--debug` is specified.
* Refactor `EventCatcher` so that the event to catch is setable
We want to be able to catch more than just `SpacesInModelNameDeprecation`
events, and in the next commit we will alter our tests to do so. Thus
instead of writing a new catcher for each event type, a slight modification
to the existing `EventCatcher` makes this much easier.
* Add project flag to control whether spaces are allowed in model names
* Log errors and raise exception when `allow_spaces_in_model_names` is `False`
* Use custom event for output invalid name counts instead of `Note` events
Using `Note` events was causing test flakiness when run in a multi
worker environment using `pytest -nauto`. This is because the event
manager is currently a global. So in a situation where test `A` starts
and test `tests_debug_when_spaces_in_name` starts shortly there after,
the event manager for both tests will have the callbacks set in
`tests_debug_when_spaces_in_name`. Then if something in test `A` fired
a `Note` event, this would affect the count of `Note` events that
`tests_debug_when_spaces_in_name` sees, causing assertion failures. By
creating a custom event, `TotalModelNamesWithSpacesDeprecation`, we limit
the possible flakiness to only tests that fire the custom event. Thus
we didn't _eliminate_ all possibility of flakiness, but realistically
only the tests in `test_check_for_spaces_in_model_names.py` can now
interfere with each other. Which still isn't great, but to fully
resolve the problem we need to work on how the event manager is
handled (preferably not globally).
* Always log total invalid model names if at least one
Previously we only logged out the count of how many invalid model names
there were if there was two or more invalid names (and not in debug mode).
However this message is important if there is even one invalid model
name and regardless of whether you are running debug mode. That is because
automated tools might be looking for the event type to track if anything
is wrong.
A related change in this commit is that we now only output the debug hint
if it wasn't run with debug mode. The idea being that if they are already
running it in debug mode, the hint could come accross as somewhat
patronizing.
* Reduce duplicate `if` logic in `check_for_spaces_in_model_names`
* Improve readability of logs related to problematic model names
We want people running dbt to be able to at a glance see warnings/errors
with running their project. In this case we are focused specifically on
errors/warnings in regards to model names containing spaces. Previously
we were only ever emitting the `warning_tag` in the message even if the
event itself was being emitted at an `ERROR` level. We now properly have
`[ERROR]` or `[WARNING]` in the message depending on the level. Unfortunately
we couldn't just look what level the event was being fired at, because that
information doesn't exist on the event itself.
Additionally, we're using events that base off of `DynamicEvents` which
unfortunately hard coded to `DEBUG`. Changing this would involve still
having a `level` property on the definition in `core_types.proto` and
then having `DynamicEvent`s look to `self.level` in the `level_tag`
method. Then we could change how firing events works based on the an
event's `level_tag` return value. This all sounds like a bit of tech
debt suited for PR, possibly multiple, and thus is not being done here.
* Alter `TotalModelNamesWithSpacesDeprecation` message to handle singular and plural
* Remove duplicate import in `test_graph.py` introduced from merging in main
---------
Co-authored-by: Emily Rockman <emily.rockman@dbtlabs.com>
* Expect that the `args` variable is un-modified by `dbt.invoke(args)`
* Make `args` variable un-modified by `dbt.invoke(args)`
* Changelog entry
* Expect that the `args` variable is un-modified by `make_dbt_context`
* Make the `args` variable is un-modified by `make_dbt_context`
* Make a copy of `args` passed to `make_dbt_context`
* Revert "Make a copy of `args` passed to `make_dbt_context`"
This reverts commit 79227b4d346b505f9f6abcbd0921074e8fe365e9.
* Ensure BaseRunner handles nodes without `build_path`
Some nodes, like SourceDefinition nodes, don't have a `build_path` property.
This is problematic because we take in nodes with no type checking, and
assume they have properties sometimes, like `build_path`. This was just
the case in BaseRunner's `_handle_generic_exception` and
`_handle_interal_exception` methods. Thus to stop dbt from crashing when
trying to handle an exception related to a node without a `build_path`,
we added an private method to the BaseRunner class for safely trying
to get `build_path`.
* Use keyword arguments when instantiating `Note` events in freshness.py
Previously we were passing arguments during the `Note` event instantiations
in freshness.py as positional arguments. This would cause not the desired
`Note` event to be emitted, but instead get the message
```
[Note] Don't use positional arguments when constructing logging events
```
which was our fault, not the users'. Additionally, we were passing the
level for the event in the `Note` instantiation when we needed to be
passing it to the `fire_event` method.
* Raise error when `loaded_at_field` is `None` and metadata check isn't possible
Previously if a source freshness check didn't have a `loaded_at_field` and
metadata source freshness wasn't supported by the adapter, then we'd log
a warning message and let the source freshness check continue. This was problematic
because the source freshness check couldn't actually continue and the process
would raise an error in the form
```
type object argument after ** must be a mapping, not NoneType
```
because the `freshness` variable was never getting set. This error wasn't particularly
helpful for any person running into it. So instead of letting that error
happen we now deliberately raise an error with helpful information.
* Add test which ensures bad source freshness checks raise appropriate error
This test directly tests that when a source freshness check doesn't have a
`loaded_at_field` and the adapter in use doesn't support metadata checks,
then the appropriate error message gets raised. That is, it directly tests
the change made in a162d53a8. This test indirectly tests the changes in both
7ec2f82a9 and 7b0ff3198 as the appropriate error can only be raised because
we've fixed other upstream issues via those commits.
* Add changelog entry for source freshness edgecase fixes
* Add @p.profile and @p.target to the list of "global" CLI flags
* Add env vars (DBT_PROFILE, DBT_TARGET) to the params
* Add unit test
* Simplify unit test
* changie
* Update .changes/unreleased/Features-20231115-092005.yaml
Co-authored-by: Doug Beatty <44704949+dbeatty10@users.noreply.github.com>
* Fix incorrect envvar names
* Realign environment variable names
* Remove from specific subcommands
* Add test_global_flags_not_on_subcommands
* Remove one unnecessary test case
* Remove other unnecessary test case
---------
Co-authored-by: Doug Beatty <doug.beatty@dbtlabs.com>
Co-authored-by: Doug Beatty <44704949+dbeatty10@users.noreply.github.com>
Our `protobuf` dep was in the section of `setup.py` which we delineate
as expecting all future versions of it to be compatible. However, this
is no longer actually the case, and in e4fe839e45
we restricted it to major vesion 4.
* [#9570] Fix fixtures in fixtures/subfolders throwing parsing error
* Fast-forward imports to match upstream
* Re-introduce doc strings on traceback info handling
* [#9570] Changelog update for fix of fixtures in fixtures/subfolders throwing parsing error
* [#9570] Improve testability and coverage for partial parsing
* Transform skip_parsing (private variable of ManifestLoader.load()) into instance-attribute of ManifestLoader(), with default value False
(to enable splitting of ManifestLoader.load())
* Split ManifestLoader.load(), to extract operation of PartialParsing into new method called ManifestLoader.safe_update_project_parser_files_partially()
(to simplify both cognitive complexity in the codebase and mocking in unittestest)
* Add "ignore" type-comments in new ManifestLoader.safe_update_project_parser_files_partially()
(to silence mypy warnings regarding instance-attributes which can be initialized as None or as something else, e.g. self.saved_manifest)[1]
[1] Although I wanted avoid "ignore" type-comments, it seems like addressing these mypy warnings in a stricter sense requires technical alignment and broader code changes.
For example, might need to initialize self.saved_manifest as Manifest, instead of Optional[Manifest], so that PartialParsing gets inputs with type it currently expects.
... perhaps too far beyond the scope of this fix?
* Check for equality with existing input_measures when adding input_measures
* Changie
* Add type annotation
* Move add_input_measure to metric from type_params
* Add tests to check that saved queries show in `dbt list`
* Update `list` task to support saved queries
This is built off of @jtcohen6 work in d6e7cda on jerco/fix-9532.
I didn't directly cherry pick because there was more work to do as
well as merge conflicts. That is to say @jtcohen6 should be credited
with some of the work.
* Update error message when iterating over nodes during list command errors
This was originally suggested by @jtcohen6 in d6e7cda of jerco/fix-9532.
This commit just makes sure the change gets included because I didn't
cherry-pick that commit into this work.
* Add test around deleting a YAML file containing semantic models and metrics
It was raised in https://github.com/dbt-labs/dbt-core/issues/8860 that an
error is being raised during partial parsing when files containing
metrics/semantic models are deleted. In further testing it looks like this
error specifically happens when a file containing both semantic models and
metrics is deleted. If the deleted file contains just semantic models or
metrics there seems to be no issue. The next commit should contain the fix.
* Skip deleted schema files when scheduling files during partial parsing
Waaaay back (in 7563b99) deleted schema files started being separated out
from deleted non-schema files. However ever since, when it came to scheduling
files for reparsing, we've only done so for deleted non-schema files. We even
missed this when we refactored the scheduling code in b37e5b5. This change
updates `_schedule_for_parsing` which is used by `schedule_nodes_for_parsing`
to begin skipping deleted schema files in addition to deleted non schema files.
* Update `add_to_pp_files` to ignore `deleted_schema_files`
As noted in the previous commit, we started separating out deleted
schema files from deleted non-schema files a looong time ago. However,
this whole time we've been adding `deleted_schema_files` to the list
of files to be parsed. This change corrects for that.
* Add changie doc for partial parsing KeyError fix
Protobuf v5 has breaking changes. Here we are limiting the protobuf
dependency to one major version, 4, so that we don't have to patch
over handling 2 different major versions of protobuf.
* Clearer no-op logging in stubbed SavedQueryRunner
* Add changelog entry
* Fix unit test
* More logging touchups
* Fix failing test
* Rename flag + refactor per #9629
* Fix failing test
* regenerate core_proto_types with libprotoc 25.3
---------
Co-authored-by: Michelle Ark <michelle.ark@dbtlabs.com>
A recent update to the version ranges for our internally
maintained support packages quite reasonably expanded the
allowed versions for dbt-semantic-interfaces to all minor versions
after 0.5.0, under the assumption that subsequent releases will
generally be backwards-compatible.
Unfortunately, dbt-semantic-interfaces is not yet in that state.
So we update the version range accordingly, and include some
comments around version range expectations for dependencies
listed in this section of dbt-core's package configuration.
CVE-2024-22195 identified an issue in Jinja2 versions <= 3.1.2. As such
we've gone and changed our dependency requirement specification to be
3.1.3 or greater (but less than 4).
Note: Preivously we were using the `~=` version specifier. However due
to some issues with the `~=` we've moved to using `>=` in combination
with `<`. This gives us the same range that `~=` gave us, but avoids
a pip resolution issue when multiple packages in an environment use `~=`
for the same dependency.
* Remove extraneous `/` in `schema-check.yml`
We have hypothesis that the extra `/` in `schema-check` is causing
issues we're seeing currently in the artifact check failing. It may
not be the final solution, but we should fix it anyway.
* Move `artifact_minor_upgrade` label check to job level of `Check Artifact Changes`
Previously the checking for `artifact_minor_upgrade` was happening in each job
step of `Check Artifact Changes`. By moving it up to the job level instead of
in the job steps we make it so the check for the label only happens once and
it simplifies the job steps.
* Update `Check Artifact Changes` to use `dorny/paths-filter`
Previously we were using `git diff` to check if any files had changed
in `core/dbt/artifacts`. However, our `git diff` usage was including any
changes that happened on `main` which the PR branch did not have. This
commit switches the check from using `git diff` to `dorny/paths-filter`,
which is what we use for checking for changelog existence as well. The
`dorny/paths-fitler` includes logic for excluding changes that are on main
but not the PR branch (which is what want to happen).
body:'Remove usage of dbt.flags.LOG_CACHE_EVENTS in dbt/adapters'
time:2023-11-01T17:31:24.974093-04:00
custom:
Author:michelleark
Issue:"8969"
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
