Hi @Evgeny
This is quite interesting scenario, what’s happen is the input/output logs become large and when there is delay the current sand-boxed will be cleaned, so the engine try to save the state but since the logs are too large, It’s try to trim it down to 2 MB which remove the large array as first step, as outside the context of delay It’s just for saving the logs in the user interface.
We have to think of tradeoff here but one thing we agree on is that user experience is so bad as it doesn’t tell you what’s happened.
To fix the issue I would recommend using subflows, extract the common part inside loop to subflow.
Or use the queue method 🔀 Queue Piece for Aggregation / Throttling / Rate Limiting