all_or_nothing transactions #590
Conversation
flumine/execution/transaction.py
Outdated
| @@ -224,5 +251,15 @@ def __enter__(self): | |||
| return self | |||
|
|
|||
| def __exit__(self, exc_type, exc_val, exc_tb): | |||
| if self.all_or_nothing: | |||
There was a problem hiding this comment.
Move to below ln264 to keep things quick
renaming all_or_nothing to atomic clearing the transaction if there's a failure.
|
Thanks for the comments. I have updated the PR accordingly. |
|
Next problem is the |
flumine/execution/transaction.py
Outdated
| @@ -146,6 +168,9 @@ def replace_order( | |||
| def execute(self) -> int: | |||
| packages = [] | |||
| if self._pending_place: | |||
| for order, market_version in self._pending_place: | |||
There was a problem hiding this comment.
We can't delay like this as it will result in the controls that act on live/open/count orders/trades incorrect
There was a problem hiding this comment.
I see that RunnerContext.place updates these fields:
- invested
- datetime_last_placed
- trades
- live_trades
I've updated the transaction to remove failed trades from live_trades, but am unsure whether the other 3 fields need to be reverted as well.
If a transaction contains multiple orders on the same runner, working out which datetime_last_placed to set it to could be tricky.
Similarly for RunnerContext.trades, if a trade contains multiple orders, but only one of them fails, which element from RunnerContext.trades should we remove?
…cute(...)." This reverts commit 4dd5ce6.
This is to address issue:
#542