Tale Distributed Transactions

What?

Tale is a small library to help write a "distributed transaction like" object across a number of services. It's loosely based on the saga pattern. A good intro is available on the couchbase blog: https://blog.couchbase.com/saga-pattern-implement-business-transactions-using-microservices-part/

Installation

pipenv install talepy

Example Usage

An example use case of this would be some holiday booking software broken down into a few services.

Assuming we have the following services: Flight booking API, Hotel booking API, and a Customer API.

We'd write the following steps:

from talepy.steps import Step

class DebitCustomerBalance(Step):

    def __init__(self):
        self.payment_client= {}

    def execute(self, state):
        state['payment_id'] = self.payment_client.bill(state.customer_id, state.payment_amount)
        return state
        
    def compensate(self, state):
        self.payment_client.refund(state['payment_id'])
       

and so on for any of the steps needed. Then in whatever is handling the user's request a distributed transaction can be built:

from talepy import run_transaction

run_transaction(
    steps=[
        DebitCustomerBalance(), 
        BookFlight(), 
        BookHotel(), 
        EmailCustomerDetailsOfBooking()
    ],
    starting_state={}
)

If any step along the way fails then the compensate method on each step is called in reverse order until everything is undone.

Steps as Lambdas

For some cases you may not want to create a class for the step. Lambdas can be used directly instead. Extending the previous example:

from talepy import run_transaction

run_transaction(
    steps=[
        DebitCustomerBalance(), 
        BookFlight(),
        lambda _: print("LOG -- The flight has been booked"),
        BookHotel(), 
        EmailCustomerDetailsOfBooking()
    ],
    starting_state={}
)

This new print statement will now execute following a success in BookFlight.

It's also possible to implement compensations by adding another lambda as a tuple pair:

from talepy import run_transaction

run_transaction(
    steps=[
        DebitCustomerBalance(), 
        BookFlight(),
        (
            lambda _: print("LOG -- The flight has been booked"), 
            lambda _: print("LOG -- ┌[ ಠ ▃ ಠ ]┐ something went wrong")
        ),
        BookHotel(), 
        EmailCustomerDetailsOfBooking()
    ],
    starting_state={}
)

Automatic retries

You may also want to try a step a few times before giving up. A continence function is provided to help out with this. Starting with the initial example. If the hotel booking step is a bit unreliable and we want to try it 3 times:

from talepy import run_transaction
from talepy.retries import attempt_retries

run_transaction(
    steps=[
        DebitCustomerBalance(), 
        BookFlight(),
        attempt_retries(BookHotel(), times=2), 
        EmailCustomerDetailsOfBooking()
    ],
    starting_state={}
)

The book hotel step will now be executed 3 times before the transaction is aborted. Once all these attempts fail the normal compensation logic will be applied.

Async

If you want to make use of async in your steps you can use run_async_transaction instead. This behaves slightly differently to run_transaction as the ordering of the steps isn't guaranteed. This means all steps receive the same starting state.

from talepy.parallel import run_async_transaction
from talepy.steps import Step

class AsyncBookFlight(Step):

    async def execute(self, state):
        # do something
        return state
        
    async def compensate(self, state):
        # revert something
        pass
       

await run_async_transaction(
    steps=[
        DebitCustomerBalance(), 
        AsyncBookFlight(),
        EmailCustomerDetailsOfBooking()
    ],
    starting_state={}
)

Testing / Development

TODO