Change log¶
All notable changes to Safir will be documented in this file.
Versioning follows semver.
This project uses scriv to maintain the change log. Changes for the upcoming release can be found in changelog.d.
9.1.1 (2024-12-18)¶
Bug fixes¶
Unset the
ssl.VERIFY_X509_STRICT
flag in SSL contexts used for Kafka connections. Python 3.13 enables this flag by default, and the current certs that Strimzi generates for these Kafka endpoints are missing an Authority Key Identifier, which prevents connections when the flag is set.
9.1.0 (2024-12-17)¶
New features¶
Add new
CountedPaginatedQueryRunner
and a returnedCountedPaginatedList
model for services that want paginated database queries that always do a second query for the total count of records without pagination.
Other changes¶
Safir is now tested with Python 3.13 as well as Python 3.12.
9.0.1 (2024-12-11)¶
Bug fixes¶
Fix dependencies on safir-logging and safir-arq from the safir PyPI module to allow the latest versions to be installed.
9.0.0 (2024-12-11)¶
Backwards-incompatible changes¶
Rewrite the Safir UWS support to use Pydantic models for job parameters. Services built on the Safir UWS library will need to change all job creation dependencies to return Pydantic models.
UWS clients must now pass an additional
job_summary_type
argument toUWSAppSettings.build_uws_config
and implementto_xml_model
in their implementation ofParametersModel
, returning a subclass of the vo-modelsParameters
class.Use the Wobbly service rather than a direct database connection to store UWS job information. Services built on the Safir UWS library must now configure a Wobbly URL and will switch to Wobbly’s storage instead of their own when updated to this release of Safir.
Case-insensitivity of form
POST
parameters to UWS routes is now handled by middleware, and theuws_post_params_dependency
function has been removed. Input parameter dependencies for UWS applications can now assume that all parameter keys will be in lowercase.Support an execution duration of 0 in the Safir UWS library, mapping it to no limit on the execution duration. Note that this will not be allowed by the default configuration and must be explicitly allowed by an execution duration validation hook.
Convert all models returned by the Safir UWS library to Pydantic. Services built on the Safir UWS library will have to change the types of validator functions for destruction time and execution duration.
Safir no longer provides the
safir.uws.ErrorCode
enum or the exceptionsafir.uws.MultiValuedParameterError
. These values were specific to a SODA service, and different IVOA UWS services use different error codes. The Safir UWS library now takes error code as a string, and each application should define its own set of error codes in accordance with the IVOA standard it is implementing.
New features¶
Add unit testing support for application metrics. Tests can define a mock application metrics event handler and inspect it after running application code to see what events would have been published.
Bug fixes¶
Append a colon after the error code when reporting UWS errors.
Other changes¶
Render all UWS XML output except for error VOTables using vo-models rather than hand-written XML templates.
8.0.0 (2024-11-26)¶
Backwards-incompatible changes¶
Add serializers to
HumanTimedelta
andSecondsTimedelta
that serialize those Pydantic fields to a float number of seconds instead of ISO 8601 durations. This means those data types now can be round-tripped (serialized and then deserialized to the original value), whereas before they could not be.parse_isodatetime
andnormalize_isodatetime
now accept exactly the date formats accepted by the IVOA DALI standard. This means seconds are now required, the trailingZ
is now optional (times are always interpreted as UTC regardless), and the time is optional and interpreted as 00:00:00 if missing.
New features¶
Add new
safir.pydantic.UtcDatetime
type that is equivalent todatetime
but coerces all incoming times to timezone-aware UTC. This type should be used instead of usingnormalize_datetime
as a validator.Add new
safir.pydantic.IvoaIsoDatetime
type that accepts any ISO 8601 date and time that matches the IVOA DALI standard for timestamps. This follows the same rules asparse_isodatetime
now follows. This type should be used instead of usingnormalize_isodatetime
as a validator.Add new
safir.database.PaginatedLinkData
model that parses the contents of an HTTPLink
header and extracts pagination information.Add
safir.database.drop_database
utility function to drop all database tables mentioned in a SQLAlchemy ORM schema and delete the Alembic version information if it is present. This is primarily useful for tests.Publishing a metrics event no longer waits on confirmation of message delivery to Kafka. This makes publishing much more performant. All events will still be delivered as long as an app awaits
EventManager.aclose
in its lifecycle.safir.kafka.PydanticSchemaManager
takes an optional structlogBoundLogger
. If not provided, the default logger is aBoundLogger
, rather than alogging.Logger
.Pass the
logger
parameter tosafir.metrics.KafkaMetricsConfiguration.make_manager
to thePydanticSchemaManager
instance that it creates.Add
CaseInsensitiveFormMiddleware
to lower-case the keys of formPOST
parameters. This can be used to support the case-insensitive keys requirement of IVOA standards.
Bug fixes¶
Correctly validate stringified floating-point numbers of seconds as inputs to the
SecondsTimedelta
type instead of truncating it to an integer.Allow
timedelta
as a member of a union field in asafir.metrics.EventPayload
.Add missing dependency on alembic to the
safir[uws]
extra.
Other changes¶
Document how to test an application’s database schema against its Alembic migrations to ensure that the schema hasn’t been modified without adding a corresponding migration.
7.0.0 (2024-11-08)¶
Backwards-incompatible changes¶
Remove
EventDuration
fromsafir.metrics
. Instead, use Python’s built-intimedelta
as the type for any field that previously usedEventDuration
. It will be serialized the same way, as an Avrodouble
number of seconds.
Bug fixes¶
Support multi-valued parameters as input to UWS
POST
routes when the values are specified by repeating the parameter.Declare that the
safir.metrics
package is typed by adding apy.typed
file.
6.5.1 (2024-10-21)¶
Bug fixes¶
Allow and ignore extra attributes to
MetricsConfiguration
when metrics are disabled. This allows passing a partial metrics configuration even when they are disabled, which simplifies the structure of Phalanx configurations based on dumping the Helm values into a YAML configuraiton file.
6.5.0 (2024-10-18)¶
New features¶
Add new
safir.kafka
module to simplify configuring and constructing Kafka clients. Included in this module issafir.kafka.PydanticSchemaManager
, which supports registering and evolving Avro schemas in the Confluent schema registry via Pydantic models. Based on @jonathansick’s kafkit work.Add new
safir.metrics
module to enable publishing app metrics events.
Bug fixes¶
Fix integration issues with vo-models in the UWS support.
6.4.0 (2024-09-16)¶
New features¶
Add support functions for using Alembic to manage database schema migrations and to verify that the current database schema matches the expectations of the running application. The support in Safir is designed to use asyncpg for all database operations, which avoids having to add a separate dependency on a sync database client.
Add Alembic schema management support to the UWS library. This support is likely temporary and will probably be replaced in the future with a standalone UWS database service.
6.3.0 (2024-08-20)¶
New features¶
safir.logging
is now available as a separate PyPI package,safir-logging
, so that it can be installed in environments where the full Safir dependency may be too heavy-weight or conflict with other packages. Packages that do depend onsafir
can and should ignore this change and continue to assume depending onsafir
will be sufficient to providesafir.logging
.Allow the database URL passed to
DatabaseSessionDependency.initialize
to be a PydanticUrl
. This simplifies logic for applications that useEnvAsyncPostgresDsn
or other Pydantic URL types for their configuration.Allow the hook URL argument to
safir.testing.slack.mock_slack_webhook
to be a PydanticSecretStr
.
Bug fixes¶
safir.logging
previously exported alogger_name
variable holding the logger name configured viasafir.logging.configure_logging
. This variable was never documented and was not intended for use outside of the library. It is no longer exported, has been renamed, and is now private to the library.Fix construction of an
Availability
object reporting problems with the UWS database layer to use the correct field names and data type for the model.
6.2.0 (2024-08-02)¶
New features¶
Add new
safir.uws
andsafir.arq.uws
modules that provide the framework of an IVOA Universal Worker Service implementation.Add new
safir.testing.uws
module that provides a mock UWS job runner for testing UWS applications.safir.arq
is now available as a separate PyPI package,safir-arq
, so that it can be installed in environments where the full Safir dependency may be too heavy-weight or conflict with other packages. Thesafir[arq]
dependency will continue to work as before (by installingsafir-arq
behind the scenes).Add new
abort_job
method to instances ofsafir.arq.ArqQueue
, which tells arq to abort a job that has been queued or in progress. To successfully abort a job that has already started, the arq worker must enable support for aborting jobs.Add new utility function
safir.arq.build_arq_redis_settings
, which constructs theRedisSettings
object used to create an arq Redis queue from a Pydantic Redis DSN.Add new
safir.arq.WorkerSettings
class that models the acceptable parameters for an arqWorkerSettings
object or class that Safir applications have needed.Add new
safir.pydantic.SecondsDatetime
andsafir.pydantic.HumanDatetime
types for use in Pydantic models. These behave the same asdatetime.timedelta
fields but use custom validation. Both support a stringified number of seconds as input, and the latter also supports the interval strings parsed bysafir.datetime.parse_timedelta
.Add new types
safir.pydantic.EnvAsyncPostgresDsn
andsafir.pydantic.EnvRedisDsn
, which validate PostgreSQL and Redis DSNs but rewrite them based on the environment variables set by tox-docker. Programs using these types for their configuration will therefore automatically honor tox-docker environment variables when running the test suite.EnvAsyncPostgresDsn
also enforces that the scheme of the DSN is compatible with asyncpg and the Safir database support.Add the decorator
safir.database.retry_async_transaction
, which retries a function or method a configurable number of times if it raises a SQLAlchemy exception from the underlying database API. This is primarily intended to retry database operations aborted due to transaction isolation.safir.database.create_database_engine
now accepts the database URL as a PydanticUrl
as well as astr
.Allow the Slack webhook URL argument to
SlackWebhookClient
andSlackRouteErrorHandler
to be given as a PydanticSecretStr
instead of astr
. This simplifies code in applications that get that value from a secret.
Other changes¶
6.1.0 (2024-07-12)¶
New features¶
Add
pull_requests
toGitHubCheckSuiteModel
to capture info about any pull requests associated with a GitHub check suite event.
6.0.0 (2024-06-10)¶
Backwards-incompatible changes¶
Drop
safir.database.create_sync_session
. This was only used by services that used Dramatiq for task management, since Dramatiq is sync-only. Services based on Safir should switch to arq and use only async database connections.Drop
DatabaseSessionDependency.override_engine
. This was used for Gafaelfawr to share a database engine across all tests for speed, but this technique breaks with current versions of pytest-asyncio and is no longer used or safe to use.
New features¶
Allow the database password to be passed to
create_database_engine
andDatabaseSessionDependency.initialize
as a PydanticSecretStr
.Add new function
safir.datetime.parse_timedelta
, which parses a human-friendly syntax for specifying time durations into a Pythondatetime.timedelta
.Add support for
gs
URLs tosafir.gcs.SignedURLService
.Support pickling of
SlackException
so that subclasses of it can be thrown by arq workers and unpickled correctly when retrieving results.
Bug fixes¶
Correctly honor the
default_queue_name
argument toRedisArqQueue.initialize
.
5.2.2 (2024-03-15)¶
Bug fixes¶
Ensure that per-request database sessions provided by
db_session_dependency
are cleaned up even if the request aborts with an uncaught exception.
5.2.1 (2024-02-19)¶
Bug fixes¶
Fix the return type of
safir.datetime.parse_isodatetime
to not includeNone
since the function never returnsNone
.
5.2.0 (2024-01-19)¶
New features¶
Add a FastAPI dependency for retrieving a Gafaelfawr delegated token from the request headers:
safir.dependencies.gafaelfawr.auth_delegated_token_dependency
.
Bug fixes¶
Rewrite
CaseInsensitiveQueryMiddleware
andXForwardedMiddleware
as pure ASGI middleware rather than using the StarletteBaseHTTPMiddleware
class. The latter seems to be behind some poor error reporting of application exceptions, has caused problems in the past due to its complexity, and is not used internally by Starlette middleware.
5.1.0 (2023-12-07)¶
New features¶
Add support for label selectors in the
list_node
method of the Kubernetes mock.
5.0.0 (2023-12-05)¶
Backwards-incompatible changes¶
Safir now depends on Pydantic v2. Python code that uses any part of Safir related to Pydantic will also need to update to Pydantic v2, since the API is significantly different. See the Pydantic migration guide for more information.
safir.pydantic.validate_exactly_one_of
is now a Pydantic model validator. It must be called withmode="after"
, since it operates in the model rather than on a raw dictionary.Remove the
GitHubAppClientFactory.create_app_client
method, which does not work with the Gidgethub API. Instead, the documentation shows how to create a JWT with theGitHubAppClientFactory
and pass it with requests.safir.github.GitHubAppClientFactory
now expects the application ID and installation ID (forcreate_installation_client
) to be of typeint
, notstr
. This appears to match what GitHub’s API returns, but not what Gidgethub expects. The ID is converted to a string when passing it to Gidgethub.
New features¶
Allow the
safir.logging.LogLevel
enum to be created from strings of any case, which will allow the logging level to be specified with any case for Safir applications that use Pydantic to validate the field.Add validated but ignored optional
propagation_policy
arguments to every delete method of the Kubernetes mock for better compatibility with the actual Kubernetes API. Previously, this argument was only accepted bydelete_namespaced_job
.All mock Kubernetes methods now accept and ignore a
_request_timeout
error for better compatibility with the Kubernetes API.Add delete, list, and watch support for persistent volume claims to the Kubernetes mock.
Bug fixes¶
safir.database.datetime_to_db
,safir.datetime.format_datetime_for_logging
, andsafir.datetime.isodatetime
now accept anydatetime
object with a time zone whose offset from UTC is 0, rather than only thedatetime.UTC
time zone object.safir.pydantic.normalize_datetime
now explicitly rejects input other than seconds since epoch or datetime objects with a validation error rather than attempting to treat the input as a datetime object and potentially throwing more obscure errors.The
_request_timeout
parameters to mock Kubernetes methods now accept a float instead of an int to more correctly match the types of kubernetes_asyncio. The mock still does not accept a tuple of timeouts.Avoid reusing the same metadata object when creating a
Pod
from aJob
. Previous versions modified thespec
part of theJob
when adding additional metadata to the childPod
.
Other changes¶
Safir is now tested with Python 3.12 as well as Python 3.11.
4.5.0 (2023-09-12)¶
New features¶
Add
list_namespaced_custom_object
with watch support to the Kubernetes mock.Add watch, field selector, and label selector support to
list_namespace
in the Kubernetes mock.
Bug fixes¶
The Kubernetes mock now correctly maintains the resource version of
Ingress
,Job
, andService
objects, since they support watches which rely on resource versions.When creating a
Pod
from aJob
in the Kubernetes mock usinggenerateName
, randomize thePod
name like Kubernetes does rather than using a fixed name. This forces tests to scan correctly for pods associated with a job. If thePod
name
orgenerateName
was explicitly configured in theJob
template, honor it.read_namespace
andlist_namespace
in the Kubernetes mock now only return namespace objects that have been explicitly created, not implicit namespaces created by creating another object without making a namespace first. This more closely matches the behavior of Kubernetes while still making it easy to use the mock in a test environment simulating a pre-existing namespace.
4.4.0 (2023-09-07)¶
New features¶
Add a
safir.click.display_help
helper function that implements ahelp
command for Click-based command-line interfaces, with support for nested subcommands.Add a new
safir.asyncio.AsyncMultiQueue
data structure, which is an asyncio multi-reader queue that delivers all messages to each reader independently.Add
read_
methods for the Kubernetes object types for which the mock providedcreate_
methods (NetworkPolicy
andPersistentVolumeClaim
).
Bug fixes¶
Fix typing of the
safir.asyncio.run_with_asyncio
decorator so that it doesn’t mask the type of the underlying function.Kubernetes objects included in events are now serialized properly using the Kubernetes camel-case field names instead of the Python snake-case names. In addition to matching Kubernetes behavior more closely, this allows a watch configured with the Kubernetes model type to deserialize the object in the
object
key of the event dictionary. The type must be passed explicitly to theWatch
constructor, since kubernetes_asyncio’s type autodetection does not work with Safir’s mock.safir.testing.kubernetes.patch_kubernetes
no longer mocks the entireApiClient
class since it is required for deserialization of objects in Kubernetes events. It instead mocks therequest
method of that class for safety, to prevent any network requests to Kubernetes clusters when Kubernetes is mocked.
Other changes¶
Safir now uses the Ruff linter instead of flake8 and isort.
4.3.1 (2023-07-17)¶
Bug fixes¶
Safir now pins the major version of all of its non-development dependencies. The impetus for this change is to prevent upgrades to Pydantic 2.x until Safir’s Pydantic models are ready for that upgrade, but a similar principle applies to other dependencies. These pins will be selectively relaxed once Safir has been confirmed to work with a new major release.
4.3.0 (2023-05-23)¶
New features¶
All
delete_*
APIs in the mock Kubernetes API now supportgrace_period_seconds
and aV1DeleteOptions
body. Both are ignored.delete_namespaced_job
in the mock Kubernetes API now requirespropagation_policy
to be passed as a keyword argument if provided, and does not require it be set. It is validated against the values recognized by Kubernetes, and if set toOrphan
, pods created by the job are not deleted.
Bug fixes¶
When reporting an error response to Slack using
SlackWebException
, put the response body in an attachment instead of a regular block, since it may be a full HTML error page. An attachment tells Slack to hide long content by default unless the viewer expands it.
4.2.2 (2023-05-17)¶
Bug fixes¶
Revert the documentation change in 4.2.1 to restore cross-references, since the docs-linkcheck failure appears to be a false positive.
Other changes¶
Safir now uses scriv to maintain its change log.
4.2.1 (2023-05-17)¶
Bug fixes¶
Fix syntax of literals in the
MockKubernetesApi
docstring.
4.2.0 (2023-05-17)¶
New features¶
Add create, delete, read, list (with watches), and (limited) patch support for
Ingress
objects to the mock Kubernetes API.Add create, delete, read, and list (with watches) support for
Job
objects to the mock Kuberntes API, and mock theBatchV1Api
.Add delete, read, and list (with watches) support for
Service
objects to the mock Kubernetes API.Add support for label selectors to
list_namespaced_pod
in the mock Kubernetes API.Add
create_namespaced_persistent_volume_claim
to the mock Kubernetes API for testing.Add support for deleting custom resources to the mock Kubernetes API.
Bug fixes¶
SlackWebException
now extracts the method and URL of the request from more httpx exceptions, and avoids exceptions when the request information is not present.
4.1.0 (2023-05-08)¶
New features¶
Add
read_*
methods forConfigMap
andResourceQuota
to the mock Kubernetes API for testing.Add
patch_namespaced_pod_status
to the mock Kubernetes API for testing. Application code is unlikely to call this, but it’s useful for test suites.The mock
list_namespaced_pod
Kubernetes API now supports watches (but be aware that all changes must be made through the API).
Bug fixes¶
Fix concurrency locking when watching namespace events in the Kubernetes testing mock. The previous logic degenerated into busy-waiting rather than correctly waiting on a condition variable.
Watches for namespace events using the mock Kubernetes API now start with the next event, not the first stored event, if
resource_version
is not set, aligning behavior with the Kubernetes API.
4.0.0 (2023-04-19)¶
Backwards-incompatible changes¶
Safir now requires a minimum Python version of 3.11.
safir.pydantic.validate_exactly_one_of
is now a Pydantic root validator instead of an individual field validator, which simplifies how it should be called.Custom Kubernetes objects are no longer copied before being stored in the Kubernetes mock by
create_namespaced_custom_object
, and no longer automatically get ametadata.uid
value. This brings handling of custom objects in line with the behavior of other mockedcreate_*
andreplace_*
APIs.All objects stored in the Kubernetes mock will be modified to set
api_version
,kind
, and (where appropriate)namespace
. If any of those values are set in the object, they are checked for correctness and rejected withAssertionError
if they don’t match the API call and its parameters.The mocked
create_*
Kubernetes APIs now use the namebody
for the parameter containing the API object, instead of using some other name specific to the kind of object. This fixes compatibility with the Python Kubernetes API that this class is mocking.
New features¶
The new
safir.redis.PydanticRedisStorage
class enables you to store Pydantic model objects in Redis. Asafir.redis.EncryptedPydanticRedisStorage
class also encrypts data in Redis. To use thesafir.redis
module, install Safir with theredis
extra (i.e.,pip install safir[redis]
).Add
safir.fastapi.ClientRequestError
base class for exceptions and the corresponding exception handlersafir.fastapi.client_request_error_handler
. These may be used together to raise structured exceptions and automatically transform them into structured HTTP errors with the correct HTTP status code and an error body compatible withsafir.models.ErrorModel
and FastAPI’s own internally-generated errors.Add
safir.slack.webhook.SlackWebException
, which is a child class ofsafir.slack.webhook.SlackException
that knows how to capture and report details from an underlying HTTPX exception.Add a
safir.testing.kubernetes.strip_none
helper function that makes it easier to compare Kubernetes objects against expected test data.Add a
get_namespace_objects_for_test
method to the Kubernetes mock to retrieve all objects (of any kind) in a namespace.Add a mock for the
list_nodes
Kubernetes API, which returns data set by a call to the newset_nodes_for_test
method.Add optional rudimentary namespace tracking to the Kubernetes mock. Namespaces can be created, deleted (which deletes all objects in the namespace), read, and listed. Explicit namespace creation remains optional; if an object is created in a namespace and that namespace does not exist, one will be implicitly created by the mock.
Add support for namespaced events (the older core API, not the new events API) to the Kubernetes mock. Newly-created pods with the default initial status post an event by default; otherwise, events must be created by calling the mocked
create_namespaced_event
API. Thelist_namespaced_event
API supports watches with timeouts.Add rudimentary support for
NetworkPolicy
,ResourceQuota
, andService
objects to the Kubernetes mock.Add a
safir.github
module for writing GitHub integrations with GidgetHub and Pydantic modeling.safir.github.GitHubAppClientFactory
helps create authenticated clients for GitHub Apps and app installations.safir.github.models
contains Pydantic models for GitHub v3 REST API resources.safir.github.webhooks
contains Pydantic models for GitHub webhook payloads.
Bug fixes¶
Stop adding the
X-Auth-Request-User
header to the OpenAPI schema for routes that useauth_dependency
orauth_logger_dependency
.
Other changes¶
The
safir.testing.kubernetes.MockKubernetesApi
mock now has rudimentary API documentation for the Kubernetes APIs that it supports.The documentation for running commands with
tox
has been updated for the new command-line syntax in tox v4.
3.8.0 (2023-03-15)¶
New features¶
Add
safir.slack.webhook.SlackWebhookClient
and accompanying models to post a structured message to Slack via an incoming webhook. Add asafir.slack.blockkit.SlackException
base class that can be used to create exceptions with supplemental metadata that can be sent to Slack as a formatted alert.Add a FastAPI route class (
safir.slack.webhook.SlackRouteErrorHandler
) that reports all uncaught exceptions to Slack using an incoming webhook.Add
safir.datetime.format_datetime_for_logging
to convert adatetime
object into an easily human-readable representation.Add
safir.testing.slack.mock_slack_webhook
and an associated mock class to mock a Slack webhook for testing.Add
microseconds=True
parameter tosafir.datetime.current_datetime
to get adatetime
object with microseconds filled in. By default,current_datetime
suppresses the microseconds since databases often cannot store them, but there are some timestamp uses, such as error reporting, that benefit from microseconds and are never stored in databases.
3.7.0 (2023-03-06)¶
New features¶
Add a
safir.testing.uvicorn.spawn_uvicorn
helper function to spawn an ASGI app under an external Uvicorn process. Normally, ASGI apps should be tested by passing the app directly to anhttpx.AsyncClient
, but in some cases (such as Selenium testing) the app needs to listen to regular HTTP requests.safir.database.initialize_database
now creates a non-default schema if the underlying SQLAlchemy model declares one.
Bug fixes¶
After
safir.logging.configure_uvicorn_logging
was called, exception tracebacks were no longer expanded. Add the missing processor to the logging stack.In
safir.logging.configure_uvicorn_logging
, send the access log to standard output instead of combining it with all other messages on standard error.
3.6.0 (2023-02-03)¶
New features¶
Add a
safir.models.ErrorLocation
enum holding valid values for the first element of theloc
array insafir.models.ErrorModel
. Due to limitations in Python typing,loc
is not type-checked against this enum, but it may be useful to applications constructing FastAPI-compatible error messages.Add
safir.datetime.current_datetime
to get a normalizeddatetime
object representing the current date and time.Add
safir.datetime.isodatetime
andsafir.datetime.parse_isodatetime
to convert to and from the most useful form of ISO 8601 dates, used by both Kubernetes and the IVOA UWS standard. Also addsafir.pydantic.normalize_isodatetime
to accept only that same format as input to a Pydantic model with adatetime
field.
3.5.0 (2023-01-12)¶
New features¶
Add new helper class
safir.gcs.SignedURLService
to generate signed URLs to Google Cloud Storage objects using workload identity. To use this class, depend onsafir[gcs]
.Add the
safir.testing.gcs
module, which can be used to mock the Google Cloud Storage API for testing. To use this module, depend onsafir[gcs]
.Add new helper class
safir.pydantic.CamelCaseModel
, which is identical topydantic.BaseModel
except with configuration added to accept camel-case keys using thesafir.pydantic.to_camel_case
alias generator and overrides ofdict
andjson
to export in camel-case by default.
3.4.0 (2022-11-29)¶
New features¶
safir.logging.configure_logging
andsafir.logging.configure_uvicorn_logging
now acceptProfile
andLogLevel
enums in addition to strings for theprofile
andlog_level
parameters. The new enums are preferred. Support for enums was added in preparation for changing the FastAPI template to usepydantic.BaseSettings
for itsConfiguration
class and validate the values of theprofile
andlog_level
configuration parameters.Add new function
safir.pydantic.normalize_datetime
, which can be used as a Pydantic validator fordatetime
fields. It also allows seconds since epoch as input, and ensures that the resultingdatetime
field in the model is timezone-aware and in UTC.Add new function
safir.pydantic.to_camel_case
, which can be used as a Pydantic alias generator when a model needs to accept attributes with camel-case names.Add new function
safir.pydantic.validate_exactly_one_of
, which can be used as a model validator to ensure that exactly one of a list of fields was set to a value other thanNone
.
Bug fixes¶
In
safir.testing.kubernetes.patch_kubernetes
, correctly apply a spec to the mock ofkubernetes_asyncio.client.ApiClient
. Due to an ordering bug in previous versions, the spec was previously a mock object that didn’t apply any constraints.
3.3.0 (2022-09-15)¶
New features¶
Add new function
safir.logging.configure_uvicorn_logging
that routes Uvicorn logging through structlog for consistent formatting. It also adds context to Uvicorn logs in the format recognized by Google’s Cloud Logging.Support the newer project metadata structure and URL naming used by
pyproject.toml
-only packages insafir.metadata.get_metadata
.
3.2.0 (2022-05-13)¶
New features¶
New support for arq, the Redis-based asyncio distributed queue package. The
safir.arq
module provides an arq client and metadata/result data classes with a mock implementation for testing. The FastAPI dependency,safir.dependencies.arq.arq_dependency
, provides a convenient way to use the arq client from HTTP handlers.
3.1.0 (2022-06-01)¶
New features¶
Add new FastAPI middleware
CaseInsensitiveQueryMiddleware
to aid in implementing the IVOA protocol requirement that the keys of query parameters be case-insensitive.
3.0.3 (2022-05-16)¶
Bug fixes¶
Correctly handle the possibility that
request.client
isNone
.
3.0.2 (2022-03-25)¶
Bug fixes¶
Fix handling of passwords containing special characters such as
@
and/
insafir.database
.
3.0.1 (2022-02-24)¶
Bug fixes¶
safir.database
retains the port in the database URL, if provided.
3.0.0 (2022-02-23)¶
Backward-incompatible changes¶
XForwardedMiddleware
no longer setsforwarded_proto
in the request state and instead directly updates the request scope so that subsequent handlers and middleware believe the request scheme is that given by anX-Forwarded-Proto
header. This fixes the scheme returned by other request methods and attributes such asurl_for
in cases where the service is behind an ingress that terminates TLS.
New features¶
Add new FastAPI dependencies
auth_dependency
andauth_logger_dependency
from thesafir.dependencies.gafaelfawr
module.auth_dependency
returns the username of the user authenticated via Gafaelfawr (pulled from theX-Auth-Request-User
header.auth_logger_dependency
returns the same logger aslogger_dependency
but with theuser
parameter bound to the username fromauth_dependency
.Add utility functions to initialize a database and create a sync or async session. The session creation functions optionally support a health check to ensure the database schema has been initialized.
Add new FastAPI dependency
db_session_dependency
that creates a task-local async SQLAlchemy session.Add utility functions
datetime_from_db
anddatetime_to_db
to convert between timezone-naive UTC datetimes stored in a database and timezone-aware UTC datetimes used elsewhere in a program.Add a
run_with_async
decorator that runs the decorated async function synchronously. This is primarily useful for decorating Click command functions (for a command-line interface) that need to make async calls.
2.4.2 (2022-01-24)¶
New features¶
Add a very basic and limited implementation of
list_namespaced_pod
tosafir.testing.kubernetes
.
2.4.1 (2022-01-14)¶
Bug fixes¶
In the Kubernetes mock API, change the expected body for
patch_namespaced_custom_object_status
to reflect the need to send a JSON patch for the entirety of the/status
path in order to work around a kubernetes_asyncio bug.
2.4.0 (2022-01-13)¶
New features¶
Add an
initialize_kubernetes
helper function to load Kubernetes configuration. Add thesafir.testing.kubernetes
module, which can be used to mock the Kubernetes API for testing. To use the new Kubernetes support, depend onsafir[kubernetes]
.
2.3.0 (2021-12-13)¶
New features¶
When logging in JSON format, use
severity
instead oflevel
for the log level for compatibility with Google Log Explorer.In the FastAPI
logger_dependency
, add log information about the incoming requst in a format compatible with Google Log Explorer.
2.2.0 (2021-11-09)¶
Bug fixes¶
Restore previous
http_client_dependency
behavior by enabling following redirects by default. This adjusts for the change of defaults in HTTPX 0.20.0.
2.1.1 (2021-10-29)¶
Other changes¶
Require structlog 21.2.0 and adjust logger configuration of exception handling for the expectations of that version.
2.1.0 (2021-08-09)¶
New features¶
Add
safir.models.ErrorModel
, which is a model of the error message format preferred by FastAPI. Using the model is not necessary but it’s helpful to reference it in API documentation to generate more complete information about the error messages.
Bug fixes¶
Mark all FastAPI dependencies as async so that FastAPI doesn’t run them in an external thread pool.
2.0.1 (2021-06-24)¶
New features¶
Defer creation of
httpx.AsyncClient
until the first time it is requested. Allow re-creation afteraclose()
, which makes it easier to write test suites.
2.0.0 (2021-06-21)¶
Backward-incompatible changes¶
As of this release, Safir is a helper library for FastAPI applications instead of aiohttp applications. Much of the library has changed. Authors of software that uses Safir should read the documentation again as part of the upgrade to FastAPI.
Included in this release is:
A FastAPI dependency to provide a structlog logger as configured by the
safir.logging
package, replacing the aiohttp middleware.A FastAPI dependency to provide a global
httpx.AsyncClient
, replacing the middleware that provided an aiohttp client.Starlette (FastAPI) middleware to parse
X-Forwarded-*
headers and update request information accordingly.safir.metadata.get_metadata
now returns a Pydantic model.
As of this release, Safir only supports Python 3.8 or later.
1.0.0 (2021-06-18)¶
Safir v1 will be the major release supporting aiohttp. Starting with Safir v2, Safir will become a support library for FastAPI applications.
This release has no significant difference from 0.2.0. The version bump is to indicate that the aiohttp support is stable.
0.2.0 (2021-06-10)¶
New features¶
configure_logging
now supports an optionaladd_timestamp
parameter (false by default) that adds a timestamp to each log message.
0.1.1 (2021-01-14)¶
Bug fixes¶
Fix duplicated log output when logging is configured multiple times.
0.1.0 (2020-02-26)¶
New features¶
The first release of Safir featuring:
safir.http
for adding anaiohttp.ClientSession
to your application.safir.logging
for configuring structlog loggers.safir.metadata
helps your gather and structure metadata about your application for publishing on metadata endpoints.safir.middleware
includes a logging middleware that adds a logger with bound context about the request to your Request object.Documentation about these features and a tutorial for starting a new application with the
roundtable_aiohttp_bot
template.