Enable replay V2 #291
Enable replay V2 #291
Conversation
| if self.durable_context._replay_schema == ReplaySchema.V1: | ||
| self.durable_context.actions.append(generation_state.actions) | ||
| else: | ||
| self.durable_context.actions[0].append(generation_state.actions) |
There was a problem hiding this comment.
Do we even generate TaskSet even more in V2?
There was a problem hiding this comment.
We do! It's just its corresponding actions field that is now represented with either a WhenAllAction or WhenAnyAction.
I did originally think of removing the TaskSet class with this PR, but that broke far too many unit tests to be viable at the moment. It is high on my priority list though
| end_time = None | ||
|
|
||
| if replay_schema == ReplaySchema.V2: | ||
| actions = WhenAllAction(actions) |
There was a problem hiding this comment.
Isn't WhenAllAction a single action, and actions expecting an Action[]?
There was a problem hiding this comment.
You're asking why the single action WhenAllAction is assigned to a variable named actions, right? Yeah that was a weird choice on my end. Looking back, it's not very readable.
The reason is that, in the case of V1 replays, actions is a list of action objects. Since we already use this variable in constructing the TaskSet below, I figured I would just override its contents for V2 and keep the remaining logic intact.
In general, a lot of this PR is designed to modify as little of the existing logic as possible, as I worried that too big of a refactor would risk introducing bugs. That being said, in this case we're talking about a specific variable name, so I don't mind just replacing the name with something like "action_payload" which works for V1 and V2.
| task4 = context.task_any([task3]) | ||
| task5 = context.task_all([task1, task2, task4]) | ||
| task6 = context.task_any([task5]) | ||
| yield task6 |
There was a problem hiding this comment.
Glad to see such a nested test case!
2 participants
This PR is a follow-up to Azure/azure-functions-durable-extension#1836
It implements the V2 replay protocol on the Python SDK. The changes are almost exclusively with respect to the replay output, not the iteration through the history.
The change is primarily to represent compound tasks (WhenAll and WhenAny) as actions themselves, instead of simply appending their argument tasks to the actions array.
By default, we continue outputting the payload expected by the legacy (V1) replay schema.