tomodachi 0.12.4

Python 3 microservice library / framework using asyncio with HTTP, websockets, RabbitMQ / AMQP and AWS SNS+SQS support.

tomodachi - a lightweight microservices library with asyncio

A Python 3 microservice library / framework using asyncio (async / await) with HTTP, websockets, RabbitMQ / AMQP and AWS SNS+SQS built-in support for event based messaging and intra-service communication.

Tomodachi is a tiny framework designed to build fast microservices listening on HTTP or communicating over event driven message buses like RabbitMQ, AMQP, AWS (Amazon Web Services) SNS+SQS, etc. It’s designed to be extendable to make use of any type of transport layer available.

Tomodachi [友達] means friends – a suitable name for microservices working together. 😻 👬 👭 👫 😻

Please note: this is a work in progress.

tomodachi is still a highly experimental project with an unregular release schedule.

Usage

tomodachi is invoked via command line interface.

Usage: tomodachi <subcommand> [options] [args]

Options:
  -h, --help             show this help message and exit
  -v, --version          print tomodachi version
  --dependency-versions  print versions of dependencies

Available subcommands:
  run <service ...> [-c <config-file ...>] [--production]
  -c, --config <files>   use json configuration files
  -l, --log <level>      specify log level
  --production           disable restart on file changes

How do I use this? (simple install using pip)

Preferrably installation should be done via pip to get the cli alias set up automatically. Locally it is recommended to install tomodachi into a virtualenv to avoid random packages into your base site-packages.

local ~$ pip install tomodachi

Getting started 🏃

Start off with import tomodachi and add a class decorated with @tomodachi.service and/or extended from the tomodachi.Service class. Name your service class and then just add functions and triggers for how to invoke them, either by HTTP requests, event messages or by timestamps / intervals.

Basic HTTP based service 🌟

Code for a simple service which would service data over HTTP.

import tomodachi


@tomodachi.service
class Service(tomodachi.Service):
    name = '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 'id = {}'.format(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 would invoke a function when messages are published on a topic exchange.

import tomodachi


@tomodachi.service
class Service(tomodachi.Service):
    name = 'example'

    # A route / topic on which the service will subscribe to via AMQP (or AWS SNS/SQS)
    @tomodachi.amqp('example.topic')
    async def example_topic_func(self, message):
        # Received message, sending same message as response on another route / topic
        await tomodachi.amqp_publish(self, message, routing_key='example.response')

Scheduling, inter-communication between services, etc. ⚡️

There are other examples available with examples of how to use services with self-invoking methods called on a specified interval or at specific times / days. Inter-communication between different services may be established using a pub-sub type with messages over AMQP or AWS SNS+SQS which is natively supported.

Run the service 😎

# cli alias is set up if installed via pip
local ~/code/service$ tomodachi run service.py

# example if cloned from repo
local ~/code/tomodachi$ python tomodachi.py run example/http_simple_service.py

Defaults to output information on stdout.

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

tomodachi/X.X.XX
October 02, 2017 - 13:38:00,481516
Quit services with <ctrl+c>.
2017-10-02 13:38:01,234 (services.service): Initializing service "example" [id: <uuid>]
2017-10-02 13:38:01,248 (transport.http): Listening [http] on http://127.0.0.1:9700/
2017-10-02 13:38:01,248 (services.service): Started service "example" [id: <uuid>]

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: Mon, 02 Oct 2017 13:38:02 GMT
id = 1234

Example of tomodachi service containerized in Docker 🐳

Great ways to run microservices are either to run them in Docker or running them serverless. Here’s an example of getting a tomodachi service up and running in Docker in no-time. The base-image (kalaspuff/python-nginx-proxy) also sets up nginx and proxies requests from port 80 to the service backend on 8080.

We’re building a container using just two small files, the Dockerfile and the actual code for the microservice, service.py.

Dockerfile

FROM kalaspuff/python-nginx-proxy:1.1.0
WORKDIR /
RUN apt-get -y update \
    && apt-get install -y build-essential=12.3 \
    && pip install tomodachi \
    && apt-get purge -y --auto-remove build-essential \
    && apt-get clean autoclean \
    && apt-get autoremove -y \
    && rm -rf /var/lib/{apt,dpkg,cache,log}/
RUN mkdir /app
WORKDIR /app
ADD service.py .
CMD tomodachi run service.py --production

service.py

import tomodachi


@tomodachi.service
class Service(tomodachi.Service):
    name = 'example'
    options = {
        'http': {
            'port': 8080
        }
    }

    @tomodachi.http('GET', r'/')
    async def index_endpoint(self, request):
        return 'friends forever!'

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

local ~/code/service$ docker build . -t tomodachi-microservice

local ~/code/service$ docker run -ti -p 31337:80 tomodachi-microservice
2017-10-02 13:38:01,234 (services.service): Initializing service "example" [id: <uuid>]
2017-10-02 13:38:01,248 (transport.http): Listening [http] on http://127.0.0.1:8080/
2017-10-02 13:38:01,248 (services.service): Started service "example" [id: <uuid>]

Making requests to the running container.

local ~$ curl http://127.0.0.1:31337/
friends forever!

Nothing more nothing less. It’s actually as easy as that.

Requirements 👍

• Python 3.5.3+, 3.6+

• aiohttp

• aiobotocore

• aioamqp

• ujson

• uvloop

License 🙋

Offered under the MIT license

Source code 🦄

The latest developer version of tomodachi is available at the GitHub repo https://github.com/kalaspuff/tomodachi

Any questions?

What is the best way to run a tomodachi service?

There is no way to tell you how to orchestrate your infrastructure. Some people may run it containerized in a Docker environment, deployed via Terraform / Nomad / Kubernetes and some may run several services on the same environment, on the same machine. There may be best practices but theres no way telling you how to orchestrate your application environment.

Personally I would currently go for a Dockerized environment with nginx proxy in front of the service to handle all the weirdness of the web, TLS, black magic and improved upgrades for WebSockets. Take a look at my kalaspuff/docker-python-nginx-proxy base-image to get your code up and running within minutes.

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 example of how to work with pub-sub for RabbitMQ via 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.

Should I run this in production?

Yes? No? There are some projects that already have live versions in production. The library is provided as is with an unregular release schedule. It’s all still highly experimental and it depends on other experimental projects, so you have to be in charge here and decide for yourself. Let me know if you do however!

Another good idea is to drop in Sentry or other exception debugging solutions, for if your invoked functions would raise unhandled exceptions.

Who built this and why?

My name is Carl Oscar Aaro [@kalaspuff] and I’m a coder from Sweden. I simply wanted to learn more about asyncio and needed a constructive off-work project to experiment with – and here we are. Nowadays I use tomodachi as a base for many smaller projects where I just want to be able to focus on the application itself, while still having the power of building distributed systems. 🎉

• https://github.com/kalaspuff

• https://www.linkedin.com/in/carloscaraaro/

Changes

0.12.4 (2018-06-24)

• Updated aioamqp to the latest version with support for Python 3.7.

• Updated service imports for improved Python 3.7 compatibility.

0.12.3 (2018-06-12)

• Improved type hinting support.

0.12.2 (2018-06-12)

• Added stubs for type hinting via tools like mypy.

0.12.1 (2018-06-07)

• Added complete support for aiohttp 3.3.x release and aiobotocore 0.9.x releases.

0.12.0 (2018-05-31)

• Improved handling of imports to allow relative imports in services and to use better error messages if the parent package is using a reserved name.

• Preparations for aiohttp 3.3.x release which deprecates some uses for custom router.

• Preparations for upcoming Python 3.7 release.

0.11.3 (2018-05-25)

• Added additional function for message validation functionality. (github: smaaland)

• Updated documentation and examples.

0.11.2 (2018-05-19)

• Improved base documentation.

• Improved and updated examples.

• Type hinting corrections for examples.

0.11.1 (2018-05-18)

• Decorators for invoker functions already decorated with for example @tomodachi.http or @tomodachi.aws_sns_sqs is now easier to implement using the @tomodachi.decorator decorator.

• Added improved exception logging from HTTP and schedule invokers also to the AWS SNS+SQS and AMQP endpoints. Unhandled exceptions are now logged as logging.exception() to the 'exception' logger.

0.11.0 (2018-05-15)

• Propagation of exceptions in invoked functions to be able to hook in exception handlers into logging. (github: 0x1EE7)

0.10.2 (2018-05-15)

• Encoding issue for Protocol Buffers messages solved. (github: smaaland).

• Support for aiobotocore 0.8.X+.

0.10.1 (2018-04-26)

• Fixes a bug for optional dependency protobuf. message_protocol imports would break unless the google.protobuf package was installed.

0.10.0 (2018-04-20)

• Base example message protocol class for using Protocol Buffers over AMQP or AWS SNS+SQS. (github: smaaland).

• Validation of event based messages via validation function specified during decoration. (github: smaaland)

• Updates to work with aiohttp 3.1.X+.

• Improved logging functionality.

• Better type hinting and linting.

0.9.5 (2018-03-16)

• More robust handling of invoking service files that aren’t a part of a Python package.

0.9.4 (2018-03-06)

• Fixes an issue affecting websocket connections where the receive function was invalidly called twice of which one time were without error handling.

0.9.3 (2018-03-06)

• Solves an error with functions for AMQP / AWS SNS+SQS functions that are used without a message_protocol class.

• Improved disconnect and reconnect to AWS SNS+SQS via aiobotocore on hot-reload and during testing.

• Improved README with event based messaging example using AMQP.

• Added the option of running schedule tasks immediately on service start. For example a function decorated by @schedule(interval=20, immediately=True) would be run immediately on service start and then every 20 seconds.

0.9.2 (2018-03-05)

• Improved error handling for bad requests (error 400) on HTTP calls.

• File watcher for hot-reload now excludes ignored directories in a more effective way to ease CPU load and for faster boot time for projects with thousands of files which should’ve been ignored.

0.9.1 (2018-03-05)

• schedule functions limits to 20 running tasks of the same function to prevent overflows in development.

• Fixes an issue where schedule tasks stopped executing if a service was hot-reloaded on code change.

• Handles websocket cancellations better if the client would close the connection before the request had been upgraded.

0.9.0 (2018-03-04)

• Updated to use aiohttp 3.X.X+ and aiobotocore 0.6.X+.

• Dropped support for Python versions below 3.5.3 as new aiohttp requires at least Python 3.5.3. Last version with support for Python 3.5.0, 3.5.1 and 3.5.2 is tomodachi 0.8.X series.

0.8.3 (2018-03-02)

• Print stack trace for outputs from schedule invoker functions tasks instead of silently catching exceptions.

• Handle close and receive errors for websockets and cleanly close already opened websockets on service exit.

0.8.2 (2018-02-28)

• Fixed broken HTTP transports due to missing colorama import.

0.8.1 (2018-02-27)

• Correction for README in 0.8.X release.

0.8.0 (2018-02-27)

• It’s now possible to specify queue_name on AWS SNS+SQS and AMQP decorators for competing queues. If not specified an automatically generated hash will be used as queue name as it worked previously.

• Fixes an issue with relative imports from within service files, which resulted in “SystemParent module ‘’ not loaded, cannot perform relative import” or “ImportError: attempted relative import with no known parent package”. (github: 0x1EE7)

• Exceptions that are subclasses of AmqpInternalServiceError and AWSSNSSQSInternalServiceError will now also work in the same way, resulting in the messages to be retried when raised.

• Service classes now have built in log functions for setting up logging to file as well as logging. They are self.log_setup('logname', level, filename) and self.log('logname', level, message).

• HTTP services will have their access log color coded when outputting to nothing else than stdout, which should be helpful in an overview during development.

0.7.0 (2018-01-27)

• Added @websocket as a decorator type for handling websockets. A function call should return two callables which will be used for receiving messages through the socket and as a way to notify about the closure of the socket.

0.6.5 (2018-01-16)

• Updated aiohttp to latest version which solves incompabilities with yarl.

0.6.4 (2018-01-15)

• Added a stricter dependency check for yarl.

0.6.3 (2018-01-12)

• Gracefully handle exceptions thrown when receiving messages from AWS SNS+SQS. For example when invalid XML data in response which causes botocore to throw a botocore.parsers.ResponseParserError.

• Updated dependencies to allow for newer version of aiohttp 2.3.X.

• Improved type hinting.

0.6.2 (2017-11-15)

• Recreate queues and resubscribe to topics if queue is removed during runtime.

0.6.1 (2017-11-15)

• Introduced new options for AWS SNS/SQS transport to use aws_endpoint_urls for sns and sqs if the user wishes to connect to other endpoints and the actual AWS endpoints, which could be useful for development and testing. The AWS SNS/SQS examples has been updated with values to reflect these options.

• Reworked timeouts and reconnects and fixed an issue in the recreate_client method which was called on server disconnects.

0.6.0 (2017-11-15)

• Stricter version control of required packages to not break installation on major/minor related updates.

• Updates to support aiohttp 2.3.X and aiobotocore 0.5.X.

0.5.3 (2017-11-08)

• Corrects issues on timeouts and server disconnects.

• Specify fixed version for aiohttp to not break installation.

• Code cleanup to conform with pycodestyle.

0.5.2 (2017-10-08)

• Add argument option for log level as ‘-l’ or ‘–log’. (github: djKooks)

• Better matching of imported modules on hot-reload which will cause reloading into code with syntax errors or indentation errors much harder.

0.5.1 (2017-10-03)

• More improvements regarding hot-reloading of code that may have syntax errors, indentation errors or issues when the service is being initiated.

0.5.0 (2017-10-02)

• Solves the issue where hot-loading into a state where the code errors due to syntax errors would crash the application, making the user need to manually restart the process.

0.4.10 (2017-10-02)

• Fixes for failing tests on hot-reloading during test phase.

0.4.9 (2017-10-02)

• Solves issue with Segmentation fault in Python 3.6 during hot-reload on Linux.

0.4.8 (2017-10-02)

• Fixes type hinting issues with Python 3.5.1.

0.4.7 (2017-09-30)

• Reworked watcher since it ended up using 90% CPU of the running core due to constant re-indexing (mstat) of every file every 0.5s. Full re-index will now only run every 10 seconds, since it’s more rare that new files are added than existing files edited. Watcher for edited existing files will still run at the same intervals.

• Watched file types may now be specified via configuration via options.watcher.watched_file_endings.

0.4.6 (2017-09-29)

• Messages via SNS+SQS or AMQP over 60000 bytes as JSON will now be sent in a gzipped base64 encoded format to allow for larger limits and lower potential SNS costs due to multiplexed messaging.

• Fixes an issue with multidict 3.2.0 on hot-reload which made the tomodachi application crash.

0.4.5 (2017-09-07)

• Possibility to requeue messages that result in specific exceptions. Exceptions that will nack the message (for AMQP transport) is called AmqpInternalServiceError. Exceptions that won’t delete the message from the queue and in turn will result in it to “reappear” unless configured non-default (for AWS SNS+SQS transport) is called AWSSNSSQSInternalServiceError.

0.4.4 (2017-08-25)

• Corrected an issue regarding crontab notation for scheduling function calls where it didn’t parse the upcoming date correctly if both isoweekday and day part were given.

0.4.3 (2017-08-09)

• Catches unintended HTTP exceptions and prints a useful stacktrace if log_level is set to DEBUG.

0.4.2 (2017-08-07)

• Fixes an issue where Content-Type header couldn’t be specified without charset in HTTP transports.

• Cleared some old debug code.

0.4.1 (2017-08-05)

• Corrects and issue with AMQP transport which caused invoked functions to not be able to declare scope variables without crashes.

0.4.0 (2017-08-05)

• Release fixes a major issue which caused invoked functions to not be able to declare any scope variables.

• @http_static decorator for serving static files from a folder on disk. Takes to values; 1. the path to the folder, either relative to the service file or absolute; 2. the base URL path for static files as a regexp.

0.3.0 (2017-07-25)

• Changed format of access log for HTTP requests - now logging user agent and login name (if authorization via Basic Auth).

• Support for X-Forwarded-For headers via real_ip_from and real_ip_header options which will log the forwarded IP instead of the one from the load balancer / proxy.

• Access log for HTTP can now be specified as a filename to which the service will log all requests.

• Fixes issue with schedule invoker which would crash if invoked at second 0.

• Updated dependencies to latest available versions.

0.2.17 (2017-07-05)

• Timezone support for schedule invoker functions.

• Added more decorator invoker functions as aliases for common scheduler use cases - @minutely, @hourly, @daily and @heartbeat (every second)

• Updated example services and better test cases.

• Updated aiohttp / aiobotocore / botocore dependencies.

0.2.16 (2017-07-02)

• Solved issues with aiobotocore / aiohttp dependencies.

• Refactored loader functions.

0.2.15 (2017-07-02)

• Corrected issue with configuration values for AWS and AWS SNS+SQS settings.

• Improved testing suite and more code coverage for integration tests.

0.2.14 (2017-06-30)

• New “transport” invoker for service functions: schedule. It works like cron type scheduling where specific functions will be run on the specified interval. For example a function can be specified to run once per day at a specific time or every second minute, or the last Tuesday of January and March at 05:30 AM.

• Values for keyword arguments invoked by transport decorators were earlier always set to None, despite having other default values. This is now corrected.

0.2.13 (2017-06-20)

• Type hinted examples and test cases.

• Shielded function calls for AMQP and SNS+SQS transports to avoid unexpected execution stop.

• Added version output to tomodachi CLI tool.

• Additional test cases.

0.2.12 (2017-06-18)

• Type hinted code base and minor bug fixes for internal functions.

0.2.11 (2017-06-09)

• Invoker methods can now be called directly without the need to mock the invoker decorator function.

0.2.10 (2017-06-08)

• Added @functools.wraps decorator to invoked functions of service classes.

0.2.9 (2017-06-06)

• Added a list of safe modules that may never be removed from the list of already loaded modules. Removing the module ‘typing’ from the list would cause a RecursionError exception since Python 3.6.1.

0.2.8 (2017-05-23)

• Additional improvements to network connectivity issues to not get stuck in waiting state.

0.2.7 (2017-05-23)

• Improved SNS+SQS draining / restart when network connectivity has been lost or temporarily suspended. Would improve situations when development machine has been in hibernation.

• Replaced deprecated logging functions to rid warnings.

0.2.6 (2017-05-22)

• Support for a “generic” aws dictonary in options that can hold region, access key id and secret to be shared among other AWS resources/services.

• Updated aiobotocore / botocore dependencies.

• Gracefully handle and discard invalid SNS/SQS messages not in JSON format.

• Corrected issue where watched directories with “similar” names as settings would be ignored.

0.2.5 (2017-05-16)

• Updated issues with function caching due to keepalive when hot reloading in development. Currently disables keepalive entirely.

• Fixed issue with updated file logging for watcher.

0.2.4 (2017-05-12)

• Downgraded botocore to meet requirements and to make the installed tomodachi script runnable again.

0.2.3 (2017-05-10)

• Watcher is now configurable to ignore specific directories dependant on the service. (github: smaaland)

• Fixed issue where using --config instead of -c would result in a raised exception. (github: smaaland)

0.2.2 (2017-05-04)

• tomodachi.transport.http has its own Response object that works better with default content types and charsets - examples/http_service.py updated.

• No automatic conversion will be tried if the returned response of an http method is of bytes type.

0.2.1 (2017-05-03)

• Improved handling of how charsets and encodings work with aiohttp.

• Fixed an issue where Content-Type header would always be included twice for aiohttp.web.Response objects.

0.2.0 (2017-05-02)

• Watcher now only reacts to files with file endings .py, .json, .yml, .html or .html and ignores to look at paths __pycache__, .git, .svn, __ignored__, __temporary__ and __tmp__.

• HTTP transport may now respond with an aiohttp.web.Response object for more complex responses.

• HTTP transport response headers can now use the multidict library.

0.1.11 (2017-04-02)

• Working PyPI release.

• Added unit tests.

• Works with aiohttp 2 and aiobotocore 0.3.

• Service classes must be decorated with @tomodachi.service.