buvar 0.44.15

Asyncio plugins, components, dependency injection and configs

This is heavily inspired by Pyramid and my daily needs to fastly create and maintain microservice like applications.

a plugin mechanic

• plugin may depend on other plugins

• plugins yield tasks to run

• a context registry serves as a store for application components created by plugins

• a dependency injection creates intermediate components

• a config source is mapped to plugin specific configuration and may be fully overridden by environment vars

• structlog boilerplate for json/tty logging

• fork the process and share bound sockets

• pytest fixtures to reduce testing boilerplate

You bootstrap like following:

from buvar import plugin

plugin.stage("some.module.with.prepare")

# some.module.with.prepare
from buvar import context, plugin

class Foo:
    ...


async def task():
    asyncio.sleep(1)


async def server():
    my_component = context.get(Foo)
    await asyncio.Future()


# there is also plugin.Teardown and plugin.Cancel
async def prepare(load: plugin.Loader):
    await load('.another.plugin')

    # create some long lasting components
    my_component = context.add(Foo())

    # you may run simple tasks
    yield task()

    # you may run server tasks
    yield server()

a components and dependency injection solution

Dependency injection relies on registered adapters, which may be a function, a method, a class, a classmethod or a generic classmthod.

Dependencies are looked up in components or may be provided via kwargs.

from buvar import di

class Bar:
    pass

class Foo:
    def __init__(self, bar: Bar = None):
        self.bar = bar

    @classmethod
    async def adapt(cls, baz: str) -> Foo:
        return Foo()

async def adapt(bar: Bar) -> Foo
    foo = Foo(bar)
    return foo


async def task():
    foo = await di.nject(Foo, baz="baz")
    assert foo.bar is None

    bar = Bar()
    foo = await di.nject(Foo, bar=bar)
    assert foo.bar is bar

async def prepare():
    di.register(Foo.adapt)
    di.register(adapt)

    yield task()

a config source

buvar.config.ConfigSource is just a dict, which merges arbitrary dicts into one. It serves as the single source of truth for application variability.

You can load a section of config values into your custom attrs class instance. ConfigSource will override values by environment variables if present.

config.toml

log_level = "DEBUG"
show_warnings = "yes"

[foobar]
some = "value"

export APP_FOOBAR_SOME=thing

import attr
import toml

from buvar import config

@attr.s(auto_attribs=True)
class GeneralConfig:
    log_level: str = "INFO"
    show_warnings: bool = config.bool_var(False)


@attr.s(auto_attribs=True)
class FoobarConfig:
   some: str


source = config.ConfigSource(toml.load('config.toml'), env_prefix="APP")

general_config = source.load(GeneralConfig)
assert general_config == GeneralConfig(log_level="DEBUG", show_warnings=True)

foobar_config = source.load(FoobarConfig, 'foobar')
assert foobar_config.some == "thing"

There is a shortcut to the above approach provided by buvar.config.Config, which requires to be subclassed from it with a distinct section attribute. If one adds a buvar.config.ConfigSource component, he will receive the mapped config in one call.

from buvar import config, plugin


@attr.s(auto_attribs=True)
class GeneralConfig(config.Config):
    log_level: str = "INFO"
    show_warnings: bool = config.bool_var(False)


@attr.s(auto_attribs=True)
class FoobarConfig(config.Config, section="foobar"):
    some: str


async def prepare(load: plugin.Loader):
    # this would by typically placed in the main CLI entry point
    source = context.add(config.ConfigSource(toml.load('config.toml'), env_prefix="APP"))

    # to provide the adapter to di, which could also be done in the main entry point
    await load(config)
    foobar_config = await di.nject(FoobarConfig)

a structlog

Just structlog boilerplate.

import sys

from buvar import log

log_config = log.LogConfig(tty=sys.stdout.isatty(), level="DEBUG")
log_config.setup()

forked process and shared sockets

You may fork your process and bind and share sockets, to leverage available CPUs e.g. for serving an aiohttp microservice.

Signals like INT, TERM, HUP are forwarded to the child processes.

import aiohttp.web
from buvar import fork, plugin, di, context
from buvar_aiohttp import AioHttpConfig


async def hello(request):
    return aiohttp.web.Response(body=b"Hello, world")


async def prepare_aiohttp(load: plugin.Loader):
    await load("buvar_aiohttp")

    app = await di.nject(aiohttp.web.Application)
    app.router.add_route("GET", "/", hello)


context.add(AioHttpConfig(host="0.0.0.0", port=5678))

fork.stage(prepare_aiohttp, forks=0, sockets=["tcp://:5678"])

pytest

There are a couple of pytest fixtures provided to get your context into a reasonable state:

buvar_config_source

A dict with arbitrary application settings.

buvar_context

The basic context staging operates on.

buvar_stage

The actual stage processing all plugins.

buvar_load

The loader to add plugins to the stage.

buvar_plugin_context

The context all plugins share, when they are prepared.

Following markers may be applied to a test

buvar_plugins(plugin, ...)

Load all plugins into plugin context.

import pytest


async def prepare():
    from buvar import context

    context.add("foobar")


@pytest.mark.asyncio
@pytest.mark.buvar_plugins("tests.test_testing")
async def test_wrapped_stage_context():
    from buvar import context, plugin

    assert context.get(str) == "foobar"
    assert context.get(plugin.Cancel)


@pytest.mark.asyncio
@pytest.mark.buvar_plugins()
async def test_wrapped_stage_context_load(buvar_load):
    from buvar import context, plugin

    await buvar_load(prepare)
    assert context.get(str) == "foobar"
    assert context.get(plugin.Cancel)