CQRS/ES commanded questions

Hi, I am working with the commanded library and will love to hear some of your views.

  1. What is the difference between a saga and a process manager? Because I sense the process manager in commanded should be named a saga as it does route messages across multiple discrete contexts/aggregates.

  2. How should I write compensating actions using commanded proc mgr?

For example, let’s say I create a wallet where monies are to be drawn out of a customer’s account. This create_wallet command is encapsulated in a proc mgr which will then trigger a corresponding withdraw_account command. What if for some reason the account table is offline? Or for some reason, the transaction falls through like if a new withdrawal policy is enacted system-wide midway?

  1. Just wondering whats the advantage of the proc mgr in commanded apart from compensating actions and workflows as aggregates can emit multiple events from a single command? If i understood correctly, I dont see much difference between the two below:

A saga that listens for a WalletCreated event to trigger a WithdrawAccount command which emits a AccountWithdrawn event:

defmodule MyProj.Sagas.CreateWallet  

  def interested?(%WalletCreated{transfer_uuid: transfer_uuid}) 
  when not is_nil(transfer_uuid) do
    {:start!, transfer_uuid}
  end

  def interested?(%AccountWithdrawn{transfer_uuid: transfer_uuid}) when not is_nil(transfer_uuid) do
    {:continue!, transfer_uuid}
  end

  def handle(%CreateWallet{}, %WalletCreated{transfer_uuid: 
  transfer_uuid, wallet_id: wallet_id, funds_balance: funds_balance, 
  currency: currency}) do
    %WithdrawAccount{
      account_id: account_id,
      transfer_uuid: transfer_uuid,
      withdrawal: Money.to_decimal(funds_balance),
      currency: Cldr.validate_currency(currency) |> elem(1)
    }
  end

  def apply(%CreateWallet{} = saga, %WalletCreated{} = event) do
    %CreateWallet{
      saga
      | transfer_uuid: event.transfer_uuid,
        account_id: event.account_id,
        destination_wallet_id: event.wallet_id,
        funds: event.funds_balance,
        currency: event.currency,
        status: :withdraw_from_account
    }
  end

  def apply(%CreateWallet{} = saga, %WalletWithdrawn{}) do
    %CreateWallet
    {
      saga
      | status: :deposit_into_wallet
    }
  end

vs a single aggregate emitting two events WalletCreated and AccountWithdrawn which will also be projected and produce the same results as above

  def execute(%Wallet{wallet_id: nil}, %CreateWallet{wallet_id: 
  wallet_id, account_id: account_id, funds_balance: funds_balance, 
  currency: currency, transfer_uuid: transfer_uuid} = _create) do
    with validated_funds_balance <- {Money.new(currency, funds_balance)}
    do
      IO.inspect(transfer_uuid)
      %WalletCreated
      {
        wallet_id: wallet_id,
        account_id: account_id,
        funds_balance: validated_funds_balance,
        currency: currency,
        transfer_uuid: transfer_uuid
      }
      %AccountWithdrawn
      {
        account_id: account_id,
        transfer_uuid: transfer_uuid,
        withdrawal: Money.to_decimal(funds_balance),
        currency: Cldr.validate_currency(currency) |> elem(1)
      }
    else
      _ -> {:error, "The wallet cannot be created at this time.}
    end
  end

Many thanks.

2 Likes

What is the difference between a saga and a process manager?

A process manager is responsible for coordinating multiple aggregates. You can use a process manager to implement the Saga pattern. This is a way of implementing long-lived transactions (spanning minutes, hours, or days) that are written as a sequence of transactions that can be interleaved. All transactions in the sequence complete successfully, or compensating transactions are run to amend a partial execution (rollback on error). The error/3 callback function in a Commanded process manager is where you can handle errors and can execute any rollback commands as applicable.

Distributed Sagas: A Protocol for Coordinating Microservices by Caitie McCaffrey is a really good talk introducing the saga pattern.

In your example since both events are produced by a single aggregate there is no need for a process manager. Emitting two events from the aggregate would be the preferred approach.

3 Likes

The talk was illuminating. Many thanks :slight_smile:

I hope my sharing helps the community here

I managed to figure out the best outcome - and my experience is that for more complex workflows - a process manager is useful. But for simply emitting multiple events without the need for compensating actions, the overhead and additional complexity of a PM/saga isn’t worth it.