-
Notifications
You must be signed in to change notification settings - Fork 26
Description
Hello, my usecase is that when my agent crashes (the app/the machine/... or the server restarts), I want to resume the crashed sessions. To find those sessions that didn't finish, after restart I could look into the checkpointer and use app-specific logic to find them.
This should be doable with SQLite checkpointer, because those results are sorted, as shown here. So going through those that failed is easy as they will be at the top and I can quickly end the search. The assumption is that timestamp~checkpoint ID.
This seems impossible with Redis, because the checkpoints are not sorted. I went through the code and don't see any hints that they would be sorted (or sortable). Consequently, I'd have to search through all checkpoints which is unacceptable.
I'm also surprised I seem to be first to notice: Is this an anti-pattern or technically impossible?