Working with contexts and resources

Every Asphalt application has at least one context: the root context. The root context is typically created by the run_application() function and passed to the root component. This context will only be closed when the application is shutting down.

Most nontrivial applications will make use of subcontexts. A subcontext is a context that has a parent context. A subcontext can make use of its parent’s resources, but the parent cannot access the resources of its children. This enables developers to create complex services that work together without risking interfering with each other.

Subcontexts can be roughly divided into two types: long lived and short lived ones. Long lived subcontexts are typically used in container components to isolate its resources from the rest of the application. Short lived subcontexts, on the other hand, usually encompass some unit of work (UOW). Examples of such UOWs are:

  • handling of a request in a network service

  • running a scheduled task

  • running a test in a test suite

Contexts are “activated” by entering them using async with Context():, and exited by leaving that block. When entered, the previous active context becomes the parent context of the new one and the new context becomes the currently active context. When the async with block is left, the previously active context once again becomes the active context. The currently active context can be retrieved using current_context().

Warning

Activating contexts in asynchronous generators can lead to corruption of the context stack. This is particularly common in asynchronous pytest fixtures because pytest helper libraries such as pytest-asyncio run the async generator using two different tasks. In such cases the workaround is to activate the context in the actual test function.

Adding resources to a context

The resource system in Asphalt exists for two principal reasons:

  • To avoid having to duplicate configuration

  • To enable sharing of pooled resources, like database connection pools

Here are a few examples of services that will likely benefit from resource sharing:

  • Database connections

  • Remote service handles

  • Serializers

  • Template renderers

  • SSL contexts

When you add a resource, you should make sure that the resource is discoverable using any abstract interface or base class that it implements. This is so that consumers of the service don’t have to care if you switch the implementation to another. For example, consider a mailer service, provided by asphalt-mailer. The library has an abstract base class for all mailers, asphalt.mailer.api.Mailer. To facilitate this loose coupling of services, it adds all its configure mailer services using the Mailer interface so that components that just need some way to send email don’t have to care what implementation was chosen in the configuration.

Resources can be added to a context in two forms: static resources and resource factories. A static resource can be any arbitrary object (except None). The same object can be added to the context under several different types, as long as the type/name combination remains unique within the same context.

A resource factory is a callable that takes a Context as an argument an returns the value of the resource. There are at least a couple reasons to use resource factories instead of static resources:

  • the resource’s lifecycle needs to be bound to the local context (example: database transactions)

  • the resource requires access to the local context (example: template renderers)

Getting resources from a context

The Context class offers a few ways to look up resources.

The first one, get_resource(), looks for a resource or resource factory matching the given type and name. If the resource is found, it returns its value.

The second one, require_resource(), works exactly the same way except that it raises ResourceNotFound if the resource is not found.

The third method, request_resource(), calls get_resource() and if the resource is not found, it waits indefinitely for the resource to be added to the context or its parents. When that happens, it calls get_resource() again, at which point success is guaranteed. This is usually used only in the components’ start() methods to retrieve resources provided by sibling components.

The order of resource lookup is as follows:

  1. search for a resource in the local context

  2. search for a resource factory in the local context and its parents and, if found, generate the local resource

  3. search for a resource in the parent contexts

Injecting resources to functions

A type-safe way to use context resources is to use dependency injection. In Asphalt, this is done by adding parameters to a function so that they have the resource type as the type annotation, and a resource() instance as the default value. The function then needs to be decorated using inject():

from asphalt.core import inject, resource

@inject
async def some_function(some_arg, some_resource: MyResourceType = resource()):
    ...

To specify a non-default name for the dependency, you can pass that name as an argument to resource():

@inject
async def some_function(some_arg, some_resource: MyResourceType = resource('alternate')):
    ...

Resources can be declared to be optional too, by using either Optional or | None (Python 3.10+ only):

@inject
async def some_function(some_arg, some_resource: Optional[MyResourceType] = resource('alternate')):
    ...  # some_resource will be None if it's not found

Restrictions:

  • The resource arguments must not be positional-only arguments

  • The resources (or their relevant factories) must already be present in the context stack (unless declared optional) when the decorated function is called, or otherwise ResourceNotFound is raised

Handling resource cleanup

Any code that adds resources to a context is also responsible for cleaning them up when the context is closed. This usually involves closing sockets and files and freeing whatever system resources were allocated. This should be done in a teardown callback, scheduled using add_teardown_callback(). When the context is closed, teardown callbacks are run in the reverse order in which they were added, and always one at a time, unlike with the Signal class. This ensures that a resource that is still in use by another resource is never cleaned up prematurely.

For example:

from asphalt.core import Component, Context


class FooComponent(Component):
    async def start(self, ctx: Context) -> None:
        service = SomeService()
        await service.start(ctx)
        ctx.add_teardown_callback(service.stop)
        ctx.add_resource(service)

There also exists a convenience decorator, context_teardown(), which makes use of asynchronous generators:

from __future__ import annotations
from collections.abc import AsyncGenerator
from asphalt.core import Component, Context, context_teardown


class FooComponent(Component):
    @context_teardown
    async def start(self, ctx: Context) -> AsyncGenerator[None, BaseException | None]:
        service = SomeService()
        await service.start(ctx)
        ctx.add_resource(service)

        yield

        # This part of the function is run when the context is closing
        service.stop()

Sometimes you may want the cleanup to know whether the context was ended because of an unhandled exception. The one use that has come up so far is committing or rolling back a database transaction. This can be achieved by passing the pass_exception keyword argument to add_teardown_callback():

class FooComponent(Component):
    async def start(self, ctx: Context) -> None:
        def teardown(exception: Optional[BaseException]):
            if exception:
                db.rollback()
            else:
                db.commit()

        db = SomeDatabase()
        await db.start(ctx)
        ctx.add_teardown_callback(teardown, pass_exception=True)
        ctx.add_resource(db)

The same can be achieved with context_teardown() by storing the yielded value:

class FooComponent(Component):
    @context_teardown
    async def start(self, ctx: Context) -> AsyncGenerator[None, BaseException | None]:
        db = SomeDatabase()
        await db.start(ctx)
        ctx.add_resource(db)

        exception = yield

        if exception:
            db.rollback()
        else:
            db.commit()

If any of the teardown callbacks raises an exception, the cleanup process will still continue, but at the end a TeardownError will be raised. This exception contains all the raised exceptions in its exceptions attribute.