💻 Microservice lib designed to ease service building using Python and asyncio, with ready to use support for HTTP + WS, AWS SNS+SQS, RabbitMQ / AMQP, middlewares, envelopes, logging, lifecycles. Extend to GraphQL, protobuf, etc.

tomodachi ⌁ a lightweight microservice lib ⌁ for Python 3

^{tomodachi [友達] means friends — 🦊🐶🐻🐯🐮🐸🐍 — a suitable name for microservices working together. ✨✨
events messaging api pubsub sns+sqs amqp http queues handlers scheduling tasks microservice tomodachi}

tomodachi is a library designed to make it easy for devs to build microservices using asyncio on Python.

Includes ready implementations to support handlers built for HTTP requests, websockets, AWS SNS+SQS and RabbitMQ / AMQP for 🚀 event based messaging, 🔗 intra-service communication and 🐶 watchdog handlers.

• HTTP request handlers (API endpoints) are sent requests via the aiohttp server library. 🪢
• Events and message handlers are hooked into a message bus, such as a queue, from for example AWS (Amazon Web Services) SNS+SQS (aiobotocore), RabbitMQ / AMQP (aioamqp), etc. 📡

Using the provided handler managers, the need for devs to interface with low-level libs directly should be lower, making it more of a breeze to focus on building the business logic. 🪄

tomodachi has a featureset to meet most basic needs, for example…

• 🦸 ⋯ Graceful termination of consumers, listeners and tasks to ensure smooth deployments.
• ⏰ ⋯ Scheduled function execution (cron notation / time interval) for building watchdog handlers.
• 🍔 ⋯ Execution middleware interface for incoming HTTP requests and received messages.
• 💌 ⋯ Simple envelope building and parsing for both receiving and publishing messages.
• 📚 ⋯ Logging support via structlog with template loggers for both "dev console" and JSON output.
• ⛑️ ⋯ Loggers and handler managers built to support exception tracing, from for example Sentry.
• 📡 ⋯ SQS queues with filter policies for SNS topic subscriptions filtering messages on message attributes.
• 📦 ⋯ Supports SQS dead-letter queues via redrive policy – infra orchestration from service optional.
• 🌱 ⋯ Designed to be extendable – most kinds of transport layers or event sources can be added.

Quicklinks to the documentation 📖

This documentation README includes information on how to get started with services, what built-in functionality exists in this library, lists of available configuration parameters and a few examples of service code.

Visit https://tomodachi.dev/ for additional documentation. 📔

• Getting started / installation

Handler types / endpoint built-ins. 🛍️

• HTTP and WebSocket endpoints
• AWS SNS+SQS event messaging
• RabbitMQ / AMQP messaging
• Scheduled functions and cron

Service options to tweak handler managers. 🛠️

• Options and configuration parameters

Use the features you need. 🌮

• Middleware functionality
• Function signature keywords
• Logging and log formatters
• OpenTelemetry instrumentation

Recommendations and examples. 🧘

• Good practices for running services in production
• Example code and template services

Please note – this library is a work in progress. 🐣

Consider tomodachi as beta software. This library follows an unregular release schedule. There may be breaking changes between 0.x versions.

Usage

tomodachi is used to execute service code via command line interface or within container images. It will be installed automatically when the package is installed in the environment.

The CLI endpoint tomodachi is then used to run services defined as tomodachi service classes.

Start a service with its class definition defined in ./service/app.py by running tomodachi run service/app.py. Finally stop the service with the keyboard interrupt <ctrl+c>.

The run command has some options available that can be specified with arguments to the CLI.

Most options can also be set as an environment variable value.

For example setting environment TOMODACHI_LOGGER=json will yield the same change to the logger as if running the service using the argument --logger json.

The value for --loop can either be set to asyncio, uvloop or auto. The uvloop value can only be used if uvloop is installed in the execution environment. Note that the default auto value will currently end up using the event loop implementation that is preferred by the Python interpreter, which in most cases will be asyncio.

Use --production to disable the file watcher that restarts the service on file changes and to hide the startup info banner.

Set the minimum log level for which the loggers will emit logs to their handlers with the --log-level option. By default the minimum log level is set to info (which includes info, warning, error and critical, resulting in only the debug log records to be filtered out).

Apply the --logger option to change the log formatter that is used by the library. The default value console is mostly suited for local development environments as it provides a structured and colorized view of log records. The console colors can be disabled by setting the env value NO_COLOR=1.

If you prefer to disable log output from the library you can use disabled (and presumably add a log handler with another implementation).

The python option isn't recommended, but available if required to use the loggers from Python's built-in logging module. Note that the built-in logging module will be used any way. as the library's loggers are both added as handlers to logging.root and has propagation of records through to logging as well.

If the template loggers from the option above doesnt' cut it or if you already have your own logger (preferably a structlog logger) and processor chain set up, you can specify a --custom-logger which will also make tomodachi use your logger set up. This is suitable also if your app is using a custom logging setup that would differ in output from what the tomodachi loggers outputs.

If your logger is initialized in for example the module yourapp.logging and the initialized (structlog) logger is aptly named logger, then use --custom-logger yourapp.logging.logger (or set as an env value TOMODACHI_CUSTOM_LOGGER=yourapp.logging.logger).

The path to the logger attribute in the module you're specifying must implement debug, info, warning, error, exception, critical and preferably also new(context: Dict[str, Any]) -> Logger (as that is what primarily will be called to create (or get) a logger).

Although non-native structlog loggers can be used as custom loggers, it's highly recommended to specify a path that has been assigned a value from structlog.wrap_logger or structlog.get_logger.

Use --opentelemetry-instrument to enable OpenTelemetry auto instrumentation of the service and libraries for which the environment has installed instrumentors.

If tomodachi is installed in the environment, using the argument --opentelemetry-instrument (or setting the TOMODACHI_OPENTELEMETRY_INSTRUMENT=1 env variable value) is mostly equivalent to starting the service using the opentelemetry-instrument CLI – OTEL distros, configurators and instrumentors will be loaded automatically and OTEL_* environment values will be processed in the same way.

Getting started 🏃

First off – installation using poetry is fully supported and battle-tested (pip works just as fine)

Install tomodachi in your preferred way, wether it be poetry, pip, pipenv, etc. Installing the distribution will give your environment access to the tomodachi package for imports as well as a shortcut to the CLI alias, which later is used to run the microservices you build.

local ~$ pip install tomodachi
> ...
> Installing collected packages: ..., ..., ..., tomodachi
> Successfully installed ... ... ... tomodachi-x.x.xx

local ~$ tomodachi --version
> tomodachi x.xx.xx

tomodachi can be installed together with a set of "extras" that will install a set of dependencies that are useful for different purposes. The extras are:

• uvloop: for the possibility to start services with the --loop uvloop option.
• protobuf: for protobuf support in envelope transformation and message serialization.
• aiodns: to use aiodns as the DNS resolver for aiohttp.
• brotli: to use brotli compression in aiohttp.
• opentelemetry: for OpenTelemetry instrumentation support.
• opentelemetry-exporter-prometheus: to use the experimental OTEL meter provider for Prometheus.

Services and their dependencies, together with runtime utilities like tomodachi, should preferably always be installed and run in isolated environments like Docker containers or virtual environments.

Building blocks for a service class and microservice entrypoint

1. import tomodachi and create a class that inherits tomodachi.Service, it can be called anything… or just Service to keep it simple.
2. Add a name attribute to the class and give it a string value. Having a name attribute isn't required, but good practice.
3. Define an awaitable function in the service class – in this example we'll use it as an entrypoint to trigger code in the service by decorating it with one of the available invoker decorators. Note that a service class must have at least one decorated function available to even be recognized as a service by tomodachi run.
4. Decide on how to trigger the function – for example using HTTP, pub/sub or on a timed interval, then decorate your function with one of these trigger / subscription decorators, which also invokes what capabilities the service initially has.

Further down you'll find a desciption of how each of the built-in invoker decorators work and which keywords and parameters you can use to change their behaviour.

Note: Publishing and subscribing to events and messages may require user credentials or hosting configuration to be able to access queues and topics.

For simplicity, let's do HTTP:

• On each POST request to /sheep, the service will wait for up to one whole second (pretend that it's performing I/O – waiting for response on a slow sheep counting database modification, for example) and then issue a 200 OK with some data.
• It's also possible to query the amount of times the POST tasks has run by doing a GET request to the same url, /sheep.
• By using @tomodachi.http an HTTP server backed by aiohttp will be started on service start. tomodachi will act as a middleware to route requests to the correct handlers, upgrade websocket connections and then also gracefully await connections with still executing tasks, when the service is asked to stop – up until a configurable amount of time has passed.

import asyncio
import random

import tomodachi


class Service(tomodachi.Service):
    name = "sleepy-sheep-counter"

    _sheep_count = 0

    @tomodachi.http("POST", r"/sheep")
    async def add_to_sheep_count(self, request):
        await asyncio.sleep(random.random())
        self._sheep_count += 1
        return 200, str(self._sheep_count)

    @tomodachi.http("GET", r"/sheep")
    async def return_sheep_count(self, request):
        return 200, str(self._sheep_count)

Run services with:

local ~/code/service$ tomodachi run service.py

Beside the currently existing built-in ways of interfacing with a service, it's possible to build additional function decorators to suit the use-cases one may have.

To give a few possible examples / ideas of functionality that could be coded to call functions with data in similar ways:

• Using Redis as a task queue with configurable keys to push or pop onto.
• Subscribing to Kinesis or Kafka event streams and act on the data received.
• An abstraction around otherwise complex functionality or to unify API design.
• As an example to above sentence; GraphQL resolver functionality with built-in tracability and authentication management, with a unified API to application devs.

Additional examples will follow with different ways to trigger functions in the service.

Of course the different ways can be used within the same class, for example the very common use-case of having a service listening on HTTP while also performing some kind of async pub/sub tasks.

Basic HTTP based service 🌟

Code for a simple service which would service data over HTTP, pretty similar, but with a few more concepts added.

import tomodachi


class Service(tomodachi.Service):
    name = "http-example"

    # Request paths are specified as regex for full flexibility
    @tomodachi.http("GET", r"/resource/(?P<id>[^/]+?)/?")
    async def resource(self, request, id):
        # Returning a string value normally means 200 OK
        return f"id = {id}"

    @tomodachi.http("GET", r"/health")
    async def health_check(self, request):
        # Return can also be a tuple, dict or even an aiohttp.web.Response
        # object for more complex responses - for example if you need to
        # send byte data, set your own status code or define own headers
        return {
            "body": "Healthy",
            "status": 200,
        }

    # Specify custom 404 catch-all response
    @tomodachi.http_error(status_code=404)
    async def error_404(self, request):
        return "error 404"

RabbitMQ or AWS SNS+SQS event based messaging service 🐰

Example of a service that calls a function when messages are published on an AMQP topic exchange.

import tomodachi


class Service(tomodachi.Service):
    name = "amqp-example"

    # The "message_envelope" attribute can be set on the service class to build / parse data.
    # message_envelope = ...

    # A route / topic on which the service will subscribe to via RabbitMQ / AMQP
    @tomodachi.amqp("example.topic")
    async def example_func(self, message):
        # Received message, fordarding the same message as response on another route / topic
        await tomodachi.amqp_publish(self, message, routing_key="example.response")

AWS SNS+SQS event based messaging service 📡

Example of a service using AWS SNS+SQS managed pub/sub messaging. AWS SNS and AWS SQS together brings managed message queues for microservices, distributed systems, and serverless applications hosted on AWS. tomodachi services can customize their enveloping functionality to both unwrap incoming messages and/or to produce enveloped messages for published events / messages. Pub/sub patterns are great for scalability in distributed architectures, when for example hosted in Docker on Kubernetes.

import tomodachi


class Service(tomodachi.Service):
    name = "aws-example"

    # The "message_envelope" attribute can be set on the service class to build / parse data.
    # message_envelope = ...

    # Using the @tomodachi.aws_sns_sqs decorator to make the service create an AWS SNS topic,
    # an AWS SQS queue and to make a subscription from the topic to the queue as well as start
    # receive messages from the queue using SQS.ReceiveMessages.
    @tomodachi.aws_sns_sqs("example-topic", queue_name="example-queue")
    async def example_func(self, message):
        # Received message, forwarding the same message as response on another topic
        await tomodachi.aws_sns_sqs_publish(self, message, topic="another-example-topic")

Scheduling, inter-communication between services, etc. ⚡️

There are other examples available with code of how to use services with self-invoking methods called on a specified interval or at specific times / days, as well as additional examples for inter-communication pub/sub between different services on both AMQP or AWS SNS+SQS as shown above. See more at the examples folder.

Run the service 😎

# cli alias is set up automatically on installation
local ~/code/service$ tomodachi run service.py

# alternatively using the tomodachi.run module
local ~/code/service$ python -m tomodachi.run service.py

Defaults to output startup banner on stdout and log output on stderr.

HTTP service acts like a normal web server.

local ~$ curl -v "http://127.0.0.1:9700/resource/1234"
# > HTTP/1.1 200 OK
# > Content-Type: text/plain; charset=utf-8
# > Server: tomodachi
# > Content-Length: 9
# > Date: Sun, 16 Oct 2022 13:38:02 GMT
# >
# > id = 1234

Getting an instance of a service

If the a Service instance is needed outside the Service class itself, it can be acquired with tomodachi.get_service. If multiple Service instances exist within the same event loop, the name of the Service can be used to get the correct one.

import tomodachi

# Get the instance of the active Service.
service = tomodachi.get_service()

# Get the instance of the Service by service name.
service = tomodachi.get_service(service_name)

Stopping the service

Stopping a service can be achieved by either sending a SIGINT <ctrl+c> or SIGTERM signal to to the tomodachi Python process, or by invoking the tomodachi.exit() function, which will initiate the termination processing flow. The tomodachi.exit() call can additionally take an optional exit code as an argument, which otherwise will default to use exit code 0.

• SIGINT signal (equivalent to using <ctrl+c>)
• SIGTERM signal
• tomodachi.exit() or tomodachi.exit(exit_code)

The process' exit code can also be altered by changing the value of tomodachi.SERVICE_EXIT_CODE, however using tomodachi.exit with an integer argument will override any previous value set to tomodachi.SERVICE_EXIT_CODE.

All above mentioned ways of initiating the termination flow of the service will perform a graceful shutdown of the service which will try to await open HTTP handlers and await currently running tasks using tomodachi's scheduling functionality as well as await tasks processing messages from queues such as AWS SQS or RabbitMQ.

Some tasks may timeout during termination according to used configuration (see options such as http.termination_grace_period_seconds) if they are long running tasks. Additionally container handlers may impose additional timeouts for how long termination are allowed to take. If no ongoing tasks are to be awaited and the service lifecycle can be cleanly terminated the shutdown usually happens within milliseconds.

Function hooks for service lifecycle changes

To be able to initialize connections to external resources or to perform graceful shutdown of connections made by a service, there's a few functions a service can specify to hook into lifecycle changes of a service.

Changes to a service settings / configuration (by for example modifying the options values) should be done in the __init__ function instead of in any of the lifecycle function hooks.

Good practice – in general, make use of the _start_service (for setting up connections) in addition to the _stop_service (to close connections) lifecycle hooks. The other hooks may be used for more uncommon use-cases.

Lifecycle functions are defined as class functions and will be called by the tomodachi process on lifecycle changes:

import tomodachi


class Service(tomodachi.Service):
    name = "example"

    async def _start_service(self):
        # The _start_service function is called during initialization,
        # before consumers or an eventual HTTP server has started.
        # It's suitable to setup or connect to external resources here.
        return

    async def _started_service(self):
        # The _started_service function is called after invoker
        # functions have been set up and the service is up and running.
        # The service is ready to process messages and requests.
        return

    async def _stopping_service(self):
        # The _stopping_service function is called the moment the
        # service is instructed to terminate - usually this happens
        # when a termination signal is received by the service.
        # This hook can be used to cancel ongoing tasks or similar.
        # Note that some tasks may be processing during this time.
        return

    async def _stop_service(self):
        # Finally the _stop_service function is called after HTTP server,
        # scheduled functions and consumers have gracefully stopped.
        # Previously ongoing tasks have been awaited for completion.
        # This is the place to close connections to external services and
        # clean up eventual tasks you may have started previously.
        return

Exceptions raised in _start_service or _started_service will gracefully terminate the service.

Example of a microservice containerized in Docker 🐳

A great way to distribute and operate microservices are usually to run them in containers or even more interestingly, in clusters of compute nodes. Here follows an example of getting a tomodachi based service up and running in Docker.

We're building the service' container image using just two small files, the Dockerfile and the actual code for the microservice, service.py. In reality a service would probably not be quite this small, but as a template to get started.

Dockerfile

FROM python:3.10-bullseye
RUN pip install tomodachi
RUN mkdir /app
WORKDIR /app
COPY service.py .
ENV PYTHONUNBUFFERED=1
CMD ["tomodachi", "run", "service.py"]

service.py

import json

import tomodachi


class Service(tomodachi.Service):
    name = "example"
    options = tomodachi.Options(
        http=tomodachi.Options.HTTP(
            port=80,
            content_type="application/json; charset=utf-8",
        ),
    )
    _healthy = True

    @tomodachi.http("GET", r"/")
    async def index_endpoint(self, request):
        # tomodachi.get_execution_context() can be used for
        # debugging purposes or to add additional service context
        # in logs or alerts.
        execution_context = tomodachi.get_execution_context()

        return json.dumps({
            "data": "hello world!",
            "execution_context": execution_context,
        })

    @tomodachi.http("GET", r"/health/?", ignore_logging=True)
    async def health_check(self, request):
        if self._healthy:
            return 200, json.dumps({"status": "healthy"})
        else:
            return 503, json.dumps({"status": "not healthy"})

    @tomodachi.http_error(status_code=400)
    async def error_400(self, request):
        return json.dumps({"error": "bad-request"})

    @tomodachi.http_error(status_code=404)
    async def error_404(self, request):
        return json.dumps({"error": "not-found"})

    @tomodachi.http_error(status_code=405)
    async def error_405(self, request):
        return json.dumps({"error": "method-not-allowed"})

Building and running the container, forwarding host's port 31337 to port 80.

local ~/code/service$ docker build . -t tomodachi-microservice
# > Sending build context to Docker daemon  9.216kB
# > Step 1/7 : FROM python:3.10-bullseye
# > 3.10-bullseye: Pulling from library/python
# > ...
# >  ---> 3f7f3ab065d4
# > Step 7/7 : CMD ["tomodachi", "run", "service.py"]
# >  ---> Running in b8dfa9deb243
# > Removing intermediate container b8dfa9deb243
# >  ---> 8f09a3614da3
# > Successfully built 8f09a3614da3
# > Successfully tagged tomodachi-microservice:latest

local ~/code/service$ docker run -ti -p 31337:80 tomodachi-microservice

Making requests to the running container.

local ~$ curl http://127.0.0.1:31337/ | jq
# {
#   "data": "hello world!",
#   "execution_context": {
#     "tomodachi_version": "x.x.xx",
#     "python_version": "3.x.x",
#     "system_platform": "Linux",
#     "process_id": 1,
#     "init_timestamp": "2022-10-16T13:38:01.201509Z",
#     "event_loop": "asyncio",
#     "http_enabled": true,
#     "http_current_tasks": 1,
#     "http_total_tasks": 1,
#     "aiohttp_version": "x.x.xx"
#   }
# }

local ~$ curl http://127.0.0.1:31337/health -i
# > HTTP/1.1 200 OK
# > Content-Type: application/json; charset=utf-8
# > Server: tomodachi
# > Content-Length: 21
# > Date: Sun, 16 Oct 2022 13:40:44 GMT
# >
# > {"status": "healthy"}

local ~$ curl http://127.0.0.1:31337/no-route -i
# > HTTP/1.1 404 Not Found
# > Content-Type: application/json; charset=utf-8
# > Server: tomodachi
# > Content-Length: 22
# > Date: Sun, 16 Oct 2022 13:41:18 GMT
# >
# > {"error": "not-found"}

It's actually as easy as that to get something spinning. The hard part is usually to figure out (or decide) what to build next.

Other popular ways of running microservices are of course to use them as serverless functions, with an ability of scaling to zero (Lambda, Cloud Functions, Knative, etc. may come to mind). Currently tomodachi works best in a container setup and until proper serverless supporting execution context is available in the library, it should be adviced to hold off and use other tech for those kinds of deployments.

Available built-ins used as endpoints 🚀

As shown, there's different ways to trigger your microservice function in which the most common ones are either directly via HTTP or via event based messaging (for example AMQP or AWS SNS+SQS). Here's a list of the currently available built-ins you may use to decorate your service functions.

HTTP endpoints:

@tomodachi.http(method, url, ignore_logging=[200])
def handler(self, request, *args, **kwargs):
    ...

Usage:

Sets up an HTTP endpoint for the specified method (GET, PUT, POST, DELETE) on the regexp url. Optionally specify ignore_logging as a dict or tuple containing the status codes you do not wish to log the access of. Can also be set to True to ignore everything except status code 500.

@tomodachi.http_static(path, url)
def handler(self, request, *args, **kwargs):
    # noop
    pass

Usage:

Sets up an HTTP endpoint for static content available as GET / HEAD from the path on disk on the base regexp url.

@tomodachi.websocket(url)
def handler(self, request, *args, **kwargs):
    async def _receive(data: Union[str, bytes]) -> None:
        ...

    async def _close() -> None:
        ...

    return _receive, _close

Usage:

Sets up a websocket endpoint on the regexp url. The invoked function is called upon websocket connection and should return a two value tuple containing callables for a function receiving frames (first callable) and a function called on websocket close (second callable). The passed arguments to the function beside the class object is first the websocket response connection which can be used to send frames to the client, and optionally also the request object.

@tomodachi.http_error(status_code)
def handler(self, request, *args, **kwargs):
    ...

Usage:

A function which will be called if the HTTP request would result in a 4XX status_code. You may use this for example to set up a custom handler on "404 Not Found" or "403 Forbidden" responses.

AWS SNS+SQS messaging:

@tomodachi.aws_sns_sqs(
    topic=None,
    competing=True,
    queue_name=None,
    filter_policy=FILTER_POLICY_DEFAULT,
    visibility_timeout=VISIBILITY_TIMEOUT_DEFAULT,
    dead_letter_queue_name=DEAD_LETTER_QUEUE_DEFAULT,
    max_receive_count=MAX_RECEIVE_COUNT_DEFAULT,
    fifo=False,
    max_number_of_consumed_messages=MAX_NUMBER_OF_CONSUMED_MESSAGES
    **kwargs,
)
def handler(self, data, *args, **kwargs):
    ...

Usage:

This would set up an AWS SQS queue, subscribing to messages on the AWS SNS topic topic (if a topic is specified), whereafter it will start consuming messages from the queue.

The competing value is used when the same queue name should be used for several services of the same type and thus "compete" for who should consume the message. Since tomodachi version 0.19.x this value has a changed default value and will now default to True as this is the most likely use-case for pub/sub in distributed architectures.

Unless queue_name is specified an auto generated queue name will be used. Additional prefixes to both topic and queue_name can be assigned by setting the options.aws_sns_sqs.topic_prefix and options.aws_sns_sqs.queue_name_prefix dict values.

AWS supports two types of queues and topics, namely standard and FIFO. The major difference between these is that the latter guarantees correct ordering and at-most-once delivery. By default, tomodachi creates standard queues and topics. To create them as FIFO instead, set fifo to True.

The max_number_of_consumed_messages setting determines how many messages should be pulled from the queue at once. This is useful if you have a resource-intensive task that you don't want other messages to compete for. The default value is 10 for standard queues and 1 for FIFO queues. The minimum value is 1, and the maximum value is 10.

The filter_policy value of specified as a keyword argument will be applied on the SNS subscription (for the specified topic and queue) as the "FilterPolicy attribute. This will apply a filter on SNS messages using the chosen "message attributes" and/or their values specified in the filter. Make note that the filter policy dict structure differs somewhat from the actual message attributes, as values to the keys in the filter policy must be a dict (object) or list (array). Example: A filter policy value of {"event": ["order_paid"], "currency": ["EUR", "USD"]} would set up the SNS subscription to receive messages on the topic only where the message attribute "event" is "order_paid" and the "currency" value is either "EUR" or "USD".

If filter_policy is not specified as an argument (default), the queue will receive messages on the topic as per already specified if using an existing subscription, or receive all messages on the topic if a new subscription is set up (default). Changing the filter_policy on an existing subscription may take several minutes to propagate. Read more about the filter policy format on AWS. https://docs.aws.amazon.com/sns/latest/dg/sns-subscription-filter-policies.html

Related to the above mentioned filter policy, the aws_sns_sqs_publish function (which is used for publishing messages) can specify "message attributes" using the message_attributes keyword argument. Values should be specified as a simple dict with keys and values. Example: {"event": "order_paid", "paid_amount": 100, "currency": "EUR"}.

The visibility_timeout value will set the queue attribute VisibilityTimeout if specified. To use already defined values for a queue (default), do not supply any value to the visibility_timeout keyword – tomodachi will then not modify the visibility timeout.

Similarly the values for dead_letter_queue_name in tandem with the max_receive_count value will modify the queue attribute RedrivePolicy in regards to the potential use of a dead-letter queue to which messages will be delivered if they have been picked up by consumers max_receive_count number of times but haven't been deleted from the queue. The value for dead_letter_queue_name should either be a ARN for an SQS queue, which in that case requires the queue to have been created in advance, or a alphanumeric queue name, which in that case will be set up similar to the queue name you specify in regards to prefixes, etc. Both dead_letter_queue_name and max_receive_count needs to be specified together, as they both affect the redrive policy. To disable the use of DLQ, use a None value for the dead_letter_queue_name keyword and the RedrivePolicy will be removed from the queue attribute. To use the already defined values for a queue, do not supply any values to the keyword arguments in the decorator. tomodachi will then not modify the queue attribute and leave it as is.

Depending on the service message_envelope (previously named message_protocol) attribute if used, parts of the enveloped data would be distributed to different keyword arguments of the decorated function. It's usually safe to just use data as an argument. You can also specify a specific message_envelope value as a keyword argument to the decorator for specifying a specific enveloping method to use instead of the global one set for the service.

If you're utilizing from tomodachi.envelope import ProtobufBase and using ProtobufBase as the specified service message_envelope you may also pass a keyword argument proto_class into the decorator, describing the protobuf (Protocol Buffers) generated Python class to use for decoding incoming messages. Custom enveloping classes can be built to fit your existing architecture or for even more control of tracing and shared metadata between services.

Encryption at rest for AWS SNS and/or AWS SQS can optionally be configured by specifying the KMS key alias or KMS key id as tomodachi service options options.aws_sns_sqs.sns_kms_master_key_id (to configure encryption at rest on the SNS topics for which the tomodachi service handles the SNS -> SQS subscriptions) and options.aws_sns_sqs.sqs_kms_master_key_id (to configure encryption at rest for the SQS queues which the service is consuming). Note that an option value set to an empty string ("") or False will unset the KMS master key id and thus disable encryption at rest. If instead an option is completely unset or set to None value no changes will be done to the KMS related attributes on an existing topic or queue. It's generally not advised to change the KMS master key id/alias values for resources currently in use. If it's expected that the services themselves, via their IAM credentials or assumed role, are responsible for creating queues and topics, these options could be desirable to use. Do not use these options if you instead are using IaC tooling to handle the topics, queues and subscriptions or that they for example are created / updated as a part of deployments. Read more at https://docs.aws.amazon.com/AWSSimpleQueueService/latest/SQSDeveloperGuide/sqs-server-side-encryption.html and https://docs.aws.amazon.com/sns/latest/dg/sns-server-side-encryption.html#sse-key-terms.

AMQP messaging (RabbitMQ):

@tomodachi.amqp(
    routing_key,
    exchange_name="amq.topic",
    competing=True,
    queue_name=None,
    **kwargs,
)
def handler(self, data, *args, **kwargs):
    ...

Usage:

Sets up the method to be called whenever a AMQP / RabbitMQ message is received for the specified routing_key. By default the 'amq.topic' topic exchange would be used, it may also be overridden by setting the options.amqp.exchange_name dict value on the service class.

The competing value is used when the same queue name should be used for several services of the same type and thus "compete" for who should consume the message. Since tomodachi version 0.19.x this value has a changed default value and will now default to True as this is the most likely use-case for pub/sub in distributed architectures.

Unless queue_name is specified an auto generated queue name will be used. Additional prefixes to both routing_key and queue_name can be assigned by setting the options.amqp.routing_key_prefix and options.amqp.queue_name_prefix dict values.

Depending on the service message_envelope (previously named message_protocol) attribute if used, parts of the enveloped data would be distributed to different keyword arguments of the decorated function. It's usually safe to just use data as an argument. You can also specify a specific message_envelope value as a keyword argument to the decorator for specifying a specific enveloping method to use instead of the global one set for the service.

If you're utilizing from tomodachi.envelope import ProtobufBase and using ProtobufBase as the specified service message_envelope you may also pass a keyword argument proto_class into the decorator, describing the protobuf (Protocol Buffers) generated Python class to use for decoding incoming messages. Custom enveloping classes can be built to fit your existing architecture or for even more control of tracing and shared metadata between services.

Scheduled functions / cron / triggered on time interval:

@tomodachi.schedule(
    interval=None,
    timestamp=None,
    timezone=None,
    immediately=False,
)
def handler(self, *args, **kwargs):
    ...

Usage:

A scheduled function invoked on either a specified interval (you may use the popular cron notation as a str for fine-grained interval or specify an integer value of seconds) or a specific timestamp. The timezone will default to your local time unless explicitly stated.

When using an integer interval you may also specify wether the function should be called immediately on service start or wait the full interval seconds before its first invokation.

@tomodachi.heartbeat
def handler(self, *args, **kwargs):
    ...

Usage:

A function which will be invoked every second.

@tomodachi.minutely
@tomodachi.hourly
@tomodachi.daily
@tomodachi.monthly
def handler(self, *args, **kwargs):
    ...

Usage:

A scheduled function which will be invoked once every minute / hour / day / month.

A word on scheduled tasks in distributed contexts: What is your use-case for scheduling function triggers or functions that trigger on an interval. These types of scheduling may not be optimal in clusters with many pods in the same replication set, as all the services running the same code will very likely execute at the same timestamp / interval (which in same cases may correlated with exactly when they were last deployed). As such these functions are quite naive and should only be used with some care, so that it triggering the functions several times doesn't incur unnecessary costs or come as a bad surprise if the functions aren't completely idempotent. To perform a task on a specific timestamp or on an interval where only one of the available services of the same type in a cluster should trigger is a common thing to solve and there are several solutions to pick from., some kind of distributed consensus needs to be reached. Tooling exists, but what you need may differ depending on your use-case. There's algorithms for distributed consensus and leader election, Paxos or Raft, that luckily have already been implemented to solutions like the strongly consistent and distributed key-value stores etcd and TiKV. Even primitive solutions such as Redis SETNX commands would work, but could be costly or hard to manage access levels around. If you're on k8s there's even a simple "leader election" API available that just creates a 15 seconds lease. Solutions are many and if you are in need, go hunting and find one that suits your use-case, there's probably tooling and libraries available to call it from your service functions.

Implementing proper consensus mechanisms and in turn leader election can be complicated. In distributed environments the architecture around these solutions needs to account for leases, decision making when consensus was not reached, how to handle crashed executors, quick recovery on master node(s) disruptions, etc.

To extend the functionality by building your own trigger decorators for your endpoints, studying the built-in invoker classes should the first step of action. All invoker classes should extend the class for a common developer experience: tomodachi.invoker.Invoker.

Function signatures - keywords with transport centric values 🪄

Function handlers, middlewares and envelopes can specify additional keyword arguments in their signatures and receive transport centric values.

The following keywords can be used across all kind of handler functions, envelopes and envelopes parsing messages. These can be used to structure apps, logging, tracing, authentication, building more advanced messaging logic, etc.

AWS SNS+SQS related values - function signature keyword arguments

^{Use the following keywords arguments in function signatures (for handlers, middlewares and envelopes used for AWS SNS+SQS messages).}

HTTP related values - function signature keyword arguments

^{Use the following keywords arguments in function signatures (for handlers and middlewares used for HTTP requests).}

Middlewares for HTTP and messaging (AWS SNS+SQS, AMQP, etc.) 🧱

Middlewares can be used to add functionality to the service, for example to add logging, authentication, tracing, build more advanced logic for messaging, unpack request queries, modify HTTP responses, handle uncaught errors, add additional context to handlers, etc.

Custom middleware functions or objects that can be called are added to the service by specifying them as a list in the http_middleware and message_middleware attribute of the service class.

from .middleware import logger_middleware

class Service(tomodachi.Service):
    name = "middleware-example"
    http_middleware = [logger_middleware]
    ...

Middlewares are invoked as a stack in the order they are specified in http_middleware or message_middleware with the first callable in the list to be called first (and then also return last).

Provided arguments to middleware functions

1. The first unbound argument of a middleware function will receive the coroutine function to call next (which would be either the handlers function or a function for the next middleware in the chain). (recommended name: func)
2. (optional) The second unbound argument of a middleware function will receive the service class object. (recommended name: service)
3. (optional) The third unbound argument of a middleware function will receive the request object for HTTP middlewares, or the message (as parsed by the envelope) for message middlewares. (recommended name: request or message)

Use the recommended names to prevent collisions with passed keywords for transport centric values that are also sent to the middleware if the keyword arguments are defined in the function signature.

Calling the handler or the next middleware in the chain

When calling the next function in the chain, the middleware function should be called as an awaitable function (await func()) and for HTTP middlewares the result should most commonly be returned.

Adding custom arguments passed on to the handler

The function can be called with any number of custom keyword arguments, which will then be passed to each following middleware and the handler itself. This pattern works a bit how contextvars can be set up, but could be useful for passing values and objects instead of keeping them in a global context.

async def logger_middleware(func: Callable[..., Awaitable], *, traceid: str = "") -> Any:
    if not traceid:
        traceid = uuid.uuid4().hex
    logger = Logger(traceid=traceid)

    # Passes the logger and traceid to following middlewares and to the handler
    return await func(logger=logger, traceid=traceid)

A middleware can only add new keywords or modify the values or existing keyword arguments (by passing it through again with the new value). The exception to this is that passed keywords for transport centric values will be ignored - their value cannot be modified - they will retain their original value.

_While a middleware can modify the values of custom keyword arguments, there is no way for a middleware to completely remove any keyword that has been added by previous middlewares._

Example of a middleware specified as a function that adds tracing to AWS SQS handlers:

This example portrays a middleware function which adds trace spans around the function, with the trace context populated from a "traceparent header" value collected from a SNS message' message attribute. The topic name and SNS message identifier is also added as attributes to the trace span.

async def trace_middleware(
    func: Callable[..., Awaitable],
    *,
    topic: str,
    message_attributes: dict,
    sns_message_id: str,
) -> None:
    ctx = TraceContextTextMapPropagator().extract(carrier=message_attributes)

    with tracer.start_as_current_span(f"SNSSQS handler '{func.__name__}'", context=ctx) as span:
        span.set_attribute("messaging.system", "AmazonSQS")
        span.set_attribute("messaging.operation", "process")
        span.set_attribute("messaging.source.name", topic)
        span.set_attribute("messaging.message.id", sns_message_id)

        try:
            # Calls the handler function (or next middleware in the chain)
            await func()
        except BaseException as exc:
            logging.getLogger("exception").exception(exc)
            span.record_exception(exc, escaped=True)
            span.set_status(StatusCode.ERROR, f"{exc.__class__.__name__}: {exc}")
            raise exc

from .middleware import trace_middleware
from .envelope import Event, MessageEnvelope

class Service(tomodachi.Service):
    name = "middleware-example"
    message_envelope: MessageEnvelope(key="event")
    message_middleware = [trace_middleware]

    @tomodachi.aws_sns_sqs("example-topic", queue_name="example-queue")
    async def handler(self, event: Event) -> None:
        ...

Example of a middleware specified as a class:

A middleware can also be specified as the object of a class, in which case the __call__ method of the object will be invoked as the middleware function. Note that bound functions such as self has to be included in the signature as it's called as a normal class function.

This class provides a simplistic basic auth implementation validating credentials in the HTTP Authorization header for HTTP requests to the service.

class BasicAuthMiddleware:
    def __init__(self, username: str, password: str) -> None:
        self.valid_credentials = base64.b64encode(f"{username}:{password}".encode()).decode()

    async def __call__(
        self,
        func: Callable[..., Awaitable[web.Response]],
        *,
        request: web.Request,
    ) -> web.Response:
        try:
            auth = request.headers.get("Authorization", "")
            encoded_credentials = auth.split()[-1] if auth.startswith("Basic ") else ""

            if encoded_credentials == self.valid_credentials:
                username = base64.b64decode(encoded_credentials).decode().split(":")[0]
                # Calls the handler function (or next middleware in the chain).
                # The handler (and following middlewares) can use username in their signature.
                return await func(username=username)
            elif auth:
                return web.json_response({"status": "bad credentials"}, status=401)

            return web.json_response({"status": "auth required"}, status=401)
        except BaseException as exc:
            try:
                logging.getLogger("exception").exception(exc)
                raise exc
            finally:
                return web.json_response({"status": "internal server error"}, status=500)

from .middleware import trace_middleware

class Service(tomodachi.Service):
    name = "middleware-example"
    http_middleware = [BasicAuthMiddleware(username="example", password="example")]

    @tomodachi.http("GET", r"/")
    async def handler(self, request: web.Request, username: str) -> web.Response:
        ...

Logging and log formatting using the tomodachi.logging module 📚

A context aware logger is available from the tomodachi.logging module that can be fetched with tomodachi.logging.get_logger() or just tomodachi.get_logger() for short.

The logger is a initiated using the popular structlog package (structlog documentation), and can be used in the same way as the standard library logger, with a few additional features, such as holding a context and logging of additional values.

The logger returned from tomodachi.get_logger() will hold the context of the current handler task or request for rich contextual log records.

To get a logger with another name than the logger set for the current context, use tomodachi.get_logger(name="my-logger").

from typing import Any

import tomodachi

class Service(tomodachi.Service):
    name = "service"

    @tomodachi.aws_sns_sqs("test-topic", queue_name="test-queue")
    async def sqs_handler(self, data: Any, topic: str, sns_message_id: str) -> None:
        tomodachi.get_logger().info("received msg", topic=topic, sns_message_id=sns_message_id)

The log record will be enriched with the context of the current handler task or request and the output should look something like this if the json formatter is used (note that the example output below has been prettified – the JSON that is actually used outputs the entire log entry on one single line):

{
    "timestamp": "2023-08-13T17:44:09.176295Z",
    "logger": "tomodachi.awssnssqs.handler",
    "level": "info",
    "message": "received msg",
    "handler": "sqs_handler",
    "type": "tomodachi.awssnssqs",
    "topic": "test-topic",
    "sns_message_id": "a1eba63e-8772-4b36-b7e0-b2f524f34bff"
}

Interactions with Python's built-in logging module

Note that the log entries are propagated to the standard library logger (as long as it wasn't filtered), in order to allow third party handler hooks to pick up records or act on them. This will make sure that integrations such a Sentry's exception tracing will work out of the box.

Similarly the tomodachi logger will also by default receive records from the standard library logger as adds a logging.root handler, so that the tomodachi logger can be used as a drop-in replacement for the standard library logger. Because of this third party modules using Python's default logging module will use the same formatter as tomodachi. Note that if logging.basicConfig() is called before the tomodachi logger is initialized, tomodachi may not be able to add its logging.root handler.

Note that when using the standard library logger directly the contextual logger won't be selected by default.

import logging

from aiohttp.web import Request, Response
import tomodachi

class Service(tomodachi.Service):
    name = "service"

    @tomodachi.http("GET", r"/example")
    async def http_handler(self, request: Request) -> Response:
        # contextual logger
        tomodachi.get_logger().info("http request")

        # these two rows result in similar log records
        logging.getLogger("service.logger").info("with logging module")
        tomodachi.get_logger("service.logger").info("with tomodachi.logging module")

        # extra fields from built in logger ends up as "extra" in log records
        logging.getLogger("service.logger").info("adding extra", extra={
            "http_request_path": request.path
        })

        return Response(body="hello world")

A GET request to /example of this service would result in five log records being emitted (as shown formatted with the json formatter). The four from the example above and the last one from the tomodachi.transport.http module.

{"timestamp": "2023-08-13T19:25:15.923627Z", "logger": "tomodachi.http.handler", "level": "info", "message": "http request", "handler": "http_handler", "type": "tomodachi.http"}
{"timestamp": "2023-08-13T19:25:15.923894Z", "logger": "service.logger", "level": "info", "message": "with logging module"}
{"timestamp": "2023-08-13T19:25:15.924043Z", "logger": "service.logger", "level": "info", "message": "with tomodachi.logging module"}
{"timestamp": "2023-08-13T19:25:15.924172Z", "logger": "service.logger", "level": "info", "message": "adding extra", "extra": {"http_request_path": "/example"}}
{"timestamp": "2023-08-13T19:25:15.924507Z", "logger": "tomodachi.http.response", "level": "info", "message": "", "status_code": 200, "remote_ip": "127.0.0.1", "request_method": "GET", "request_path": "/example", "http_version": "HTTP/1.1", "response_content_length": 11, "user_agent": "curl/7.88.1", "handler_elapsed_time": "0.00135s", "request_time": "0.00143s"}

Configuring the logger

Start the service using the --logger json arguments (or setting TOMODACHI_LOGGER=json environment value) to change the log formatter to use the json log formatter. The default log formatter console is mostly suited for local development environments as it provides a structured and colorized view of log records.

It's also possible to use your own logger implementation by specifying --custom-logger ... (or setting TOMODACHI_CUSTOM_LOGGER=... environment value).

Read more about how to start the service with another formatter or implementation in the usage section

Using OpenTelemetry instrumentation

Install tomodachi using the opentelemetry extras to enable instrumentation for OpenTelemetry. In addition, install with the opentelemetry-exporter-prometheus extras to use Prometheus exporter metrics.

local ~$ pip install tomodachi[opentelemetry]
local ~$ pip install tomodachi[opentelemetry,opentelemetry-exporter-prometheus]

When added as a Poetry dependency the opentelemetry extras can be enabled by adding tomodachi = {extras = ["opentelemetry"]} to the pyproject.toml file, and when added to a requiements.txt file the opentelemetry extras can be enabled by adding tomodachi[opentelemetry] to the file.

Auto instrumentation: tomodachi --opentelemetry-instrument

Passing the --opentelemetry-instrument argument to tomodachi run will automatically instrument the service with the appropriate exporters and configuration according to the set OTEL_* environment variables.

If tomodachi is installed in the environment, using tomodachi --opentelemetry-instrument service.py is mostly equivalent to running opentelemetry-instrument tomodachi run service.py and will load distros, configurators and instrumentors automatically in the same way as the opentelemetry-instrument CLI would do.

local ~$ OTEL_LOGS_EXPORTER=console \
    OTEL_TRACES_EXPORTER=console \
    OTEL_METRICS_EXPORTER=console \
    OTEL_SERVICE_NAME=example-service \
    tomodachi --opentelemetry-instrument run service/app.py

The environment variable TOMODACHI_OPENTELEMETRY_INSTRUMENT if set will also enable auto instrumentation in the same way.

local ~$ OTEL_LOGS_EXPORTER=console \
    OTEL_TRACES_EXPORTER=console \
    OTEL_METRICS_EXPORTER=console \
    OTEL_SERVICE_NAME=example-service \
    TOMODACHI_OPENTELEMETRY_INSTRUMENT=1 \
    tomodachi run service/app.py

Auto instrumentation using the opentelemetry-instrument CLI

Auto instrumentation using the opentelemetry-instrument CLI can be achieved by starting services using opentelemetry-instrument [otel-options] tomodachi run [options] <service.py ...>.

# either define the OTEL_* environment variables to specify instrumentation specification
local ~$ OTEL_LOGS_EXPORTER=console \
    OTEL_TRACES_EXPORTER=console \
    OTEL_METRICS_EXPORTER=console \
    OTEL_SERVICE_NAME=example-service \
    opentelemetry-instrument tomodachi run service/app.py

# or use the arguments passed to the opentelemetry-instrument command
local ~$ opentelemetry-instrument \
    --logs_exporter console \
    --traces_exporter console \
    --metrics_exporter console \
    --service_name example-service \
    tomodachi run service/app.py

Manual instrumentation

Auto instrumentation using either tomodachi --opentelemetry-instrument, setting the TOMODACHI_OPENTELEMETRY_INSTRUMENT=1 env value or using the opentelemetry-instrument CLI are the recommended ways of instrumenting services, as they will automatically instrument the service (and libs with instrumentors installed) with the appropriate exporters and configuration.

However, instrumentation can also be enabled by importing the TomodachiInstrumentor instrumentation class and calling its' instrument function.

import tomodachi
from tomodachi.opentelemetry import TomodachiInstrumentor

TomodachiInstrumentor().instrument()

class Service(tomodachi.Service):
    name = "example-service"

    @tomodachi.http(GET, r"/example")
    async def example(self, request):
        return 200, "hello world"

Starting such a service with the appropriate OTEL_* environment variables would properly instrument traces, logs and metrics for the service without the need to use the opentelemetry-instrument CLI.

local ~$ OTEL_LOGS_EXPORTER=console \
    OTEL_TRACES_EXPORTER=console \
    OTEL_METRICS_EXPORTER=console \
    OTEL_SERVICE_NAME=example-service \
    tomodachi run service/app.py

Service name dynamically set if missing OTEL_SERVICE_NAME value

If the OTEL_SERVICE_NAME environment variable value (or --service_name argument to opentelemetry-instrument) is not set, the resource' service.name will instead be set to the name attribute of the service class. In case the service class uses the default generic names (service or app), the resource' service.name will instead be set to the default as specified in https://github.com/open-telemetry/semantic-conventions/tree/main/docs/resource#service.

Note for use cases where multiple service classes are started in the same Python process. Instrumentation for logging and metrics will currently primarily use the OTEL_SERVICE_NAME, and if it's missing then use the name from the first instrumented service class. The same goes for the service.instance.id resource attribute, which will be set to the first instrumented service class' uuid value (which in most cases is automatically assigned on service start).

Exclude lists to exclude certain URLs from traces and metrics

To exclude certain URLs from traces and metrics, set the environment variable OTEL_PYTHON_TOMODACHI_EXCLUDED_URLS (or OTEL_PYTHON_EXCLUDED_URLS to cover all instrumentations) to a string of comma delimited regexes that match the URLs.

Regexes from the OTEL_PYTHON_AIOHTTP_EXCLUDED_URLS environment variable will also be excluded.

For example,

export OTEL_PYTHON_TOMODACHI_EXCLUDED_URLS="client/.*/info,healthcheck"

will exclude requests such as https://site/client/123/info and https://site/xyz/healthcheck.

You can also pass comma delimited regexes directly to the instrument method:

TomodachiInstrumentor().instrument(excluded_urls="client/.*/info,healthcheck")

Prometheus meter provider (experimental)

The tomodachi.opentelemetry module also provides a Prometheus meter provider that can be used to export metrics to Prometheus. Run opentelemetry-instrument with the --meter_provider tomodachi_prometheus argument (or set OTEL_PYTHON_METER_PROVIDER=tomodachi_prometheus environment value) to enable the Prometheus meter provider.

Environment variables to configure Prometheus meter provider

• OTEL_PYTHON_TOMODACHI_PROMETHEUS_ADDRESS specifies the host address the Prometheus export server should listen on. (default: "localhost")
• OTEL_PYTHON_TOMODACHI_PROMETHEUS_PORT specifies the port the Prometheus export server should listen on. (default: 9464)
• OTEL_PYTHON_TOMODACHI_PROMETHEUS_INCLUDE_SCOPE_INFO specifies whether to include scope information as otel_scope_info value. (default: true)
• OTEL_PYTHON_TOMODACHI_PROMETHEUS_INCLUDE_TARGET_INFO specifies whether to include resource attributes as target_info value. (default: true)
• OTEL_PYTHON_TOMODACHI_PROMETHEUS_EXEMPLARS_ENABLED specifies whether exemplars (experimental) should be collected and used in Prometheus export. (default: false)
• OTEL_PYTHON_TOMODACHI_PROMETHEUS_NAMESPACE_PREFIX specifies the namespace prefix for Prometheus metrics. A final underscore is automatically added if prefix is used. (default: "")

Dependency requirement for Prometheus meter provider

The tomodachi_prometheus meter provider requires that the opentelemetry-exporter-prometheus``and ``prometheus_client packages package are installed.

Use tomodachi extras opentelemetry-exporter-prometheus to automatically include a compatible version of the exporter.

OpenMetrics output from Prometheus with exemplars enabled

With exemplars enabled, make sure to call the Prometheus client with the accept header application/openmetrics-text to ensure exemplars are included in the response.

curl http://localhost:9464/metrics -H "Accept: application/openmetrics-text"

💡 Note that if the accept header application/openmetrics-text is missing from the request, exemplars will be excluded from the response.

Example: starting a service with instrumentation

This example will start and instrument a service with OTLP exported traces sent to the endpoint otelcol:4317 and metrics that can be scraped by Prometheus from port 9464. All metrics except for target_info and otel_scope_info will be prefixed with "tomodachi_". Additionally exemplars will be added to the Prometheus collected metrics that includes sample exemplars with trace_id and span_id labels.

local ~$ TOMODACHI_OPENTELEMETRY_INSTRUMENT=1 \
    OTEL_TRACES_EXPORTER=otlp \
    OTEL_EXPORTER_OTLP_ENDPOINT=otelcol:4317 \
    OTEL_PYTHON_METER_PROVIDER=tomodachi_prometheus \
    OTEL_PYTHON_TOMODACHI_PROMETHEUS_EXEMPLARS_ENABLED=true \
    OTEL_PYTHON_TOMODACHI_PROMETHEUS_ADDRESS=0.0.0.0 \
    OTEL_PYTHON_TOMODACHI_PROMETHEUS_PORT=9464 \
    OTEL_PYTHON_TOMODACHI_PROMETHEUS_NAMESPACE_PREFIX=tomodachi \
    tomodachi run service/app.py

Additional configuration options 🤩

In the service class an attribute named options (as a tomodachi.Options object) can be set for additional configuration.

import json

import tomodachi

class Service(tomodachi.Service):
    name = "http-example"
    options = tomodachi.Options(
        http=tomodachi.Options.HTTP(
            port=80,
            content_type="application/json; charset=utf-8",
            real_ip_from=[
                "127.0.0.1/32",
                "10.0.0.0/8",
                "172.16.0.0/12",
                "192.168.0.0/16",
            ],
            keepalive_timeout=5,
            max_keepalive_requests=20,
        ),
        watcher=tomodachi.Options.Watcher(
            ignored_dirs=["node_modules"],
        ),
    )

    @tomodachi.http("GET", r"/health")
    async def health_check(self, request):
        return 200, json.dumps({"status": "healthy"})

    # Specify custom 404 catch-all response
    @tomodachi.http_error(status_code=404)
    async def error_404(self, request):
        return json.dumps({"error": "not-found"})

Decorated functions using @tomodachi.decorator 🎄

Invoker functions can of course be decorated using custom functionality. For ease of use you can then in turn decorate your decorator with the the built-in @tomodachi.decorator to ease development. If the decorator would return anything else than True or None (or not specifying any return statement) the invoked function will not be called and instead the returned value will be used, for example as an HTTP response.

import tomodachi


@tomodachi.decorator
async def require_csrf(instance, request):
    token = request.headers.get("X-CSRF-Token")
    if not token or token != request.cookies.get("csrftoken"):
        return {
            "body": "Invalid CSRF token",
            "status": 403
        }


class Service(tomodachi.Service):
    name = "example"

    @tomodachi.http("POST", r"/create")
    @require_csrf
    async def create_data(self, request):
        # Do magic here!
        return "OK"

Good practices for running services in production 🤞

When running a tomodachi service in a production environment, it's important to ensure that the service is set up correctly to handle the demands and constraints of a live system. Here's some recommendations of options and operating practices to make running the services a breeze.

• Go for a Docker 🐳 environment if possible – preferably orchestrated with for example Kubernetes to handle automated scaling events to meet demand of incoming requests and/or event queues.

• Make sure that a SIGTERM signal is passed to the python process when a pod is scheduled for termination to give it time to gracefully stop listeners, consumers and finish active handler tasks.

  • This should work automatically for services in Docker if the CMD statement in your Dockerfile is starting the tomodachi service directly.
  • In case shell scripts are used in CMD you might need to trap signals and forward them to the service process.

• To give services the time to gracefully complete active handler executions and shut down, make sure that the orchestration engine waits at least 30 seconds from sending the SIGTERM to remove the pod.

  • For extra compatibility in k8s and to get around most kind of edge-cases of intermittent timeouts and problems with ingress connections, set the pod spec terminationGracePeriodSeconds to 90 seconds and use a preStop lifecycle hook of 20 seconds.

    spec:
      terminationGracePeriodSeconds: 90
      containers:
        lifecycle:
          preStop:
            exec:
              command: ["/bin/sh", "-c", "sleep 20"]

• If your service inbound network access to HTTP handlers from users or API clients, then it's usually preferred to put some kind of ingress (nginx, haproxy or other type of load balancer) to proxy connections to the service pods.

  • Let the ingress handle public TLS, http2 / http3, client facing keep-alives and WebSocket protocol upgrades and let the service handler just take care of the business logic.

  • Use HTTP options such as the ones in this service to have the service rotate keep-alive connections so that ingress connections doesn't stick to the old pods after a scaling event.

    If keep-alive connections from ingresses to services stick for too long, the new replicas added when scaling out won't get their balanced share of the requests and the old pods will continue to receive most of the requests.

    import tomodachi
    
    class Service(tomodachi.Service):
        name = "service"
    
        options = tomodachi.Options(
            http=tomodachi.Options.HTTP(
                port=80,
                content_type="application/json; charset=utf-8",
                real_ip_from=["127.0.0.1/32", "10.0.0.0/8", "172.16.0.0/12", "192.168.0.0/16"],
                keepalive_timeout=10,
                max_keepalive_time=30,
            )
        )

• Use a JSON log formatter such as the one enabled via --logger json (or env variable TOMODACHI_LOGGER=json) so that the log entries can be picked up by a log collector.

• Always start the service with the --production CLI argument (or set the env variable TOMODACHI_PRODUCTION=1) to disable the file watcher that restarts the service on file changes, and to hide the start banner so it doesn't end up in log buffers.

• Not related to tomodachi directly, but always remember to collect the log output and monitor your instances or clusters.

Arguments to tomodachi run when running in production env

tomodachi run service/app.py --loop uvloop --production --log-level warning --logger json

Here's a breakdown of the arguments and why they would be good for these kinds of environments.

• --loop uvloop: This argument sets the event loop implementation to uvloop, which is known to be faster than the default asyncio loop. This can help improve the performance of your service. However, you should ensure that uvloop is installed in your environment before using this option.

• --production: This argument disables the file watcher that restarts the service on file changes and hides the startup info banner. This is important in a production environment where you don't want your service to restart every time a file changes. It also helps to reduce unnecessary output in your logs.

• --log-level warning: This argument sets the minimum log level to warning. In a production environment, you typically don't want to log every single detail of your service's operation. By setting the log level to warning, you ensure that only important messages are logged.

  If your infrastructure supports rapid collection of log entries and you see a clear benefit of including logs of log level info, it would make sense to use --log-level info instead of filtering on at least warning.

• --logger json: This argument sets the log formatter to output logs in JSON format. This is useful in a production environment where you might have a log management system that can parse and index JSON logs for easier searching and analysis.

You can also set these options using environment variables. This can be useful if you're deploying your service in a containerized environment like Docker or Kubernetes, where you can set environment variables in your service's configuration. Here's how you would set the same options using environment variables:

export TOMODACHI_LOOP=uvloop
export TOMODACHI_PRODUCTION=1
export TOMODACHI_LOG_LEVEL=warning
export TOMODACHI_LOGGER=json

tomodachi run service/app.py

By using environment variables, you can easily change the configuration of your service without having to modify your code or your command line arguments. This can be especially useful in a CI/CD pipeline where you might want to adjust your service's configuration based on the environment it's being deployed to.

Requirements 👍

• Python (3.8+, 3.9+, 3.10+, 3.11+)
• aiohttp (aiohttp is the currently supported HTTP server implementation for tomodachi)
• aiobotocore and botocore (used for AWS SNS+SQS pub/sub messaging)
• aioamqp (used for RabbitMQ / AMQP pub/sub messaging)
• structlog (used for logging)
• uvloop (optional: alternative event loop implementation)

Pull requests and bug reports

This library is open source software. Please add a pull request with the feature that you deem are missing from the lib or for bug fixes that you encounter.

Make sure that the tests and linters are passing. A limited number of tests can be run locally without external services. Use GitHub Actions to run the full test suite and to verify linting and regressions. Read more in the contribution guide.

GitHub repository

The latest developer version of tomodachi is always available at GitHub.

• Clone repo: git clone [email protected]:kalaspuff/tomodachi.git
• GitHub: https://github.com/kalaspuff/tomodachi

Acknowledgements + contributors

🙇 Thank you everyone that has come with ideas, reported issues, built and operated services, helped debug and made contributions to the library code directly or via libraries that build on the base functionality.

🙏 Many thanks to the amazing contributors that have helped to make tomodachi better.

Changelog of releases

Changes are recorded in the repo as well as together with the GitHub releases.

• In repository: https://github.com/kalaspuff/tomodachi/blob/master/CHANGES.rst
• Release tags: https://github.com/kalaspuff/tomodachi/releases

LICENSE

tomodachi is offered under the MIT license.

Additional questions and information

What is the best way to run a tomodachi service?

Docker containers are great and can be scaled out in Kubernetes, Nomad or other orchestration engines. Some may instead run several services on the same environment, on the same machine if their workloads are smaller or more consistent. Remember to gather your output and monitor your instances or clusters.

See the section on good practices for running services in production environments for more insights.

Are there any more example services?

There are a few examples in the examples folder, including using tomodachi in an example Docker environment with or without docker-compose. There are examples to publish events / messages to an AWS SNS topic and subscribe to an AWS SQS queue. There's also a similar code available of how to work with pub/sub for RabbitMQ via the AMQP transport protocol.

Why should I use this?

tomodachi is a perfect place to start when experimenting with your architecture or trying out a concept for a new service. It may not have all the features you desire and it may never do, but I believe it's great for bootstrapping microservices in async Python.

I have some great additions!

Sweet! Please open a pull request with your additions. Make sure that the tests and linters are passing. A limited number of tests can be run locally without external services. Use GitHub Actions to run the full test suite and to verify linting and regressions. Get started at the short contribution guide.

Beta software in production?

There are some projects and organizations that already are running services based on tomodachi in production. The library is provided as is with an unregular release schedule, and as with most software, there will be unfortunate bugs or crashes. Consider this currently as beta software (with an ambition to be stable enough for production). Would be great to hear about other use-cases in the wild!

Another good idea is to drop in Sentry or other exception debugging solutions. These are great to catch errors if something wouldn't work as expected in the internal routing or if your service code raises unhandled exceptions.

Who built this and why?

My name is Carl Oscar Aaro [@kalaspuff] and I'm a coder from Sweden. When I started writing the first few lines of this library back in 2016, my intention was just to learn more about Python's asyncio, the event loop, event sourcing and message queues. A lot has happened since – now running services in both production and development clusters, while also using microservices for quick proof of concepts and experimentation. 🎉