A Python library for writing (micro)services and their clients
Project description
A general-purpose library for writing Python (micro)services and their clients, based on an RPC (remote procedure call) calling style. Provides both a client and a server, which can be used directly by themselves or, as we do, extended with extra functionality (our authentication, metrics, and other code is all written as middleware and run on top of this library).
PySOA uses the concept of “transports” to define a layer for sending requests and responses (messages) between clients and servers. The intended transport is a Redis pub-sub layer, which we use in combination with Redis Sentinel in clusters. There is also a local transport implementation for testing and other uses.
The basic tenets of the framework are:
Services and actions both have simple names, and are called from the client by name. You can call actions individually, or bundle multiple action calls into a Job to be run serially (either aborting or continuing on error).
Requests and responses are simply Python dicts, and PySOA uses our open source validation framework conformity in order to verify their schema on the way in and out.
Message bodies are encoded using MessagePack by default (however, you can define your own serializer), with a few non-standard types encoded using msgpack’s ext, such as dates, times, date-times, and amounts of currency (using our open source currint library)
Requests have a context, which is sourced from the original client context (web request, API request, etc.) and automatically chained down into subsequent client calls made inside the service. This is used for things like correlation IDs, locales, etc.
We include “SOA Switches” as a first-party implementation of feature flags/toggles. Like the context, they are bundled along with every request and automatically chained, and are packed to try and ensure they have minimal overhead.
This intro summarizes some of the key concepts of using PySOA. For more thorough documentation, see the PySOA documentation.
Servers
SOA servers run as standalone processes and connect out to their transport to service requests and send responses, with no listening ports. This means they can easily be scaled by simply launching or killing instances with whatever orchestration software you want to use.
You can run all of the servers under a single channel layer (Redis instance/Sentinel cluster), have a separate layer per service, or have separate layers for different quality of service levels for your site based on the access point and type of accessing user.
Servers declare one or more Actions, which are registered on the class. Actions are callable objects of some type (such as a function or method, or a class with a __call__ method that will get instantiated before being called) that get called with a request and return a response. We provide a base Action class that extends this contract to also implement validation on requests and responses, but there is no requirement to use this if your needs are more complex. Actions that are classes will be passed a reference to the server’s settings object when instantiated.
from pysoa import server
from example_service.actions.call_service import CallServiceAction
from example_service.actions.square import SquareAction
from example_service.actions.status import StatusAction
class Server(server.BaseServer):
service_name = 'example'
action_class_map = {
'call_service': CallServiceAction,
'square': SquareAction,
'status': StatusAction,
}A fully-functional Example Service is available for your analysis and experimentation. We encourage you to browse its source code, and even start it up, to see how it works and get a better idea how to build services using PySOA.
Clients
Clients are instantiated with a dictionary of service names and the transports by which they can be reached. There are several approaches for calling service actions with a Client object:
Calling a single action and getting the action response back directly using call_action:
action_response = client.call_action('example', 'square', {'number': 42})Creating a single job of multiple action requests, and sending it off to all be processed by the same server instance, serially:
job_response = client.call_actions('example', [ {'action': 'square', 'body': {'number': 42}}, {'action': 'status', 'body': {'verbose': True}}, ])Creating multiple jobs, one for each action belonging to the same service, and send them off to be processed by multiple server instances in parallel:
action_responses = client.call_actions_parallel('example', [ {'action': 'square', 'body': {'number': 1035}}, {'action': 'status', 'body': {'verbose': True}}, ])Creating multiple jobs, each with its own service name and one or more actions, and send them off to be processed by multiple server instances in parallel:
job_responses = client.call_jobs_parallel([ {'service_name': 'example', 'actions': [ {'action': 'square', 'body': {'number': 4}}, {'action': 'square', 'body': {'number': 8}}, {'action': 'square', 'body': {'number': 17}}, ]}, {'service_name': 'example', 'actions': [{'action': 'status', 'body': {'verbose': True}}]}, {'service_name': 'flight_booking', 'actions': [ {'action': 'get_available_flights', 'body': { 'departure_airport': 'BNA', 'arrival_airport': 'SFO', 'departure_date': '2018-07-15', 'return_date': '2018-07-20', }}, ]}, ])
Middleware
Both clients and servers can be extended using middleware, which, in the Django style, is code that wraps around a request-response call, either on the client or server side, to add or mutate things in the request or response.
For example, some of our internal server middleware:
Reads authentication tokens from the request and validates them to make sure the request is valid and not too old
Logs metrics at the start and end of an action being processed so we can track how long our code is taking to run
Catches errors in server code and logs it into Sentry so we can track and fix problems in production
Settings
Both client and server use a dict-based settings system, with a conformity-defined schema to ensure that whatever settings are provided are valid (this schema is extensible by service implementations if they have special settings they need set).
The server also has an integration mode with Django where it will read its settings from django.conf.settings.SOA_SERVER_SETTINGS for both running and for tests, which allows easy integration of Django models and application logic into services (we make heavy use of the Django ORM in our services).
Testing
Services can be tested using standard unit tests and either by calling the actions directly (after all, they are just callable objects), or, if a run through the server machinery is desired, using the ServerTestCase base class, which takes care of setting up local transports for you.
For entire-system integration tests, you will need to spin up a copy of each desired service individually and point them at an integration-test-specific channel layer to ensure isolation from the rest of the system.
There is also a StubClient available for testing code that calls services, but where you do not actually want to have the service code in place, and a stub_action decorator / context manager that makes easy work of using it.
For more information about using these test utilities in your services or service-calling applications, see the testing documentation in the PySOA documentation.
For testing the PySOA library directly, you must first install Lua on your system (on Mac OS X this is done with brew install lua), ensure Lua is on your $PKG_CONFIG_PATH environment variable (in Mac OS X), and then install dependencies (pip install -e .[testing]). After this, you can simply run pytest or setup.py test.
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
File details
Details for the file pysoa-0.38.2.tar.gz.
File metadata
- Download URL: pysoa-0.38.2.tar.gz
- Upload date:
- Size: 97.6 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 | 33f85a623fcb68e951f0b7f68a975ddbeaee86d35e3bab7be08bf58f360cbabd |
|
| MD5 | a5b8ccc7393d55b4469966f6593742cb |
|
| BLAKE2b-256 | 1f0cf9a984a52b4da492a9a0ae46cf9fb45ab6da6b8b7755a63a1198a21f5af7 |
File details
Details for the file pysoa-0.38.2-py27.py35.py36-none-any.whl.
File metadata
- Download URL: pysoa-0.38.2-py27.py35.py36-none-any.whl
- Upload date:
- Size: 127.6 kB
- Tags: Python 2.7, Python 3.5, Python 3.6
- Uploaded using Trusted Publishing? No
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 | eb7f60bd48cef6e3ce49ea83f633989379abb72ce5b419991cc243d19fa94bef |
|
| MD5 | 08721373e7e1878c9e64526335a9eef8 |
|
| BLAKE2b-256 | 6c0d98781c28920dfb4b08e465cf3edcfdf2e0f3077b4bda1d893a900755e358 |
