Remote procedures calls (RPCs) are defined as methods on your API classes. They are useful when either:
- You require information from a service
- You wish to wait until a remote procedure has completed an action
RPCs provide at-most-once delivery semantics. If you need at-least-once semantics you should consder using events instead.
As covered in previous sections, you define RPCs as follows:
# bus.py from lightbus import Api class AuthApi(Api): class Meta: name = 'auth' def check_password(self, username, password): return username == 'admin' and password == 'secret' def reset_password(self, username): reset_users_password_somehow(username) def get_user(self, username): return get_user(username) def promote_to_admin(username): user = get_user(username) user.admin = True user.save()
RPCs are called simply as follows:
# Anywhere in your code # Import your project's bus instance from bus import bus # Call the RPC is_valid = bus.auth.check_password(username="adam", password="secr3t")
You can also perform the call asynchronously using asyncio:
# Anywhere in your code # Import your project's bus instance from bus import bus is_valid = await bus.auth.check_password.call_async( username="adam", password="secr3t" )
See the typing reference.
It is best to keep your RPCs simple and easy to understand. In anything but the simplest service it will probably be best to use the API definition as a presentational layer which wraps up the business logic located elsewhere.
If you find business logic creeping into your RPC definitions, consider factoring it out and invoking it from the RPC definition.
For smaller services this will be less important, but as functionality is shared and used elsewhere within your service you may find it keeps your code more managable.
This also leaves your RPCs definitions free do any API-specific legwork such as data marshalling. For example, converting incoming natural keys (e.g. usernames) into the primary keys (i.e. user IDs) which your service's may use internally.
Architecture & coupling¶
RPCs often represent a tight coupling between your code and the service you are calling. This may be acceptable to you, but it is worth being aware of potential pitfalls:
- Failures & timeouts may occurr, which should ideally be handled gracefully
- Modifications to the remote RPC may require updates to the code which calls the RPC
- RPCs incur much greater overhead than regular function calls. Utility functions that use RPCs should therefore make it clear that they will be incurring this overhead (either through naming convention or documentation)
- The more services call to for a given action, the less reliable the action will typically be. (i.e. if any service is unavailable the action will potentially fail)
Using events will provide a different set of trade-offs, and ultimately you will need to decide on what is right for your particular scenario.
When deciding the values your RPC should receive, consider:
- Can this value become out of date? For example, an entire user object could become out of date, whereas a username or user ID would not.
- Packing and unpacking large data structures is computationally expensive.
- Will the meaning of the value change over time? For example, the meaning of 'today' or 'now' will change, but the meaning of a specific date & time will remain the same.
RPCs can only be called with keyword arguments. For example:
# Raises an InvalidParameters exception result = bus.auth.check_password('admin', 'secret') # ERROR # Success result = bus.auth.check_password(username='admin', password='secret')