0
votes

We would like to use iteration in conjunction with the Async IO operator to execute sequential API calls for the same event. However, in an answer to one other question that I asked, it was mentioned that sing Iteration with Datastreams is a bad idea.

Manage state with huge memory usage - querying from storage

Could someone explain further?

1
This answer -- stackoverflow.com/a/54707931/2000823 -- explains the problem, which is that checkpointing isn't guaranteed to work correctly in graphs with iterations. - David Anderson

1 Answers

2
votes

While iterations itself currently does not checkpointing, there are other options, like loop unrolling (if there is a max number of iterations per element) and statefun, which uses a feedback mechanism to do achieve the same thing that iterations do but in a much more checkpoint-friendly way.