feat: bandits #11
feat: bandits #11
Conversation
|
Rust tests will fail depending on Eppo-exp/sdk-test-data#44 |
| let assignment = self | ||
| .get_assignment( | ||
| flag_key, | ||
| subject_key, | ||
| &subject_attributes.to_generic_attributes(), | ||
| Some(VariationType::String), | ||
| ) | ||
| .unwrap_or_default() | ||
| .unwrap_or_else(|| Assignment { | ||
| value: AssignmentValue::String(default_variation.to_owned()), | ||
| event: None, | ||
| }); |
There was a problem hiding this comment.
One deviation from other SDKs: because of how eppo_core is structured, it is possible to evaluate assignments without triggering assignment logging. I therefore removed an early return in case of a bandit flag with empty actions and evaluate the assignment immediately.
- In case assignment resolves to non-bandit variation, we return the variation and log the assignment.
- If it's a bandit variation and actions are missing, bandit evaluation will fail and we'll return default variation and skip assignment logging.
- Otherwise, we return variation+action + log the assignment and bandit action.
There was a problem hiding this comment.
I think you could mirror the other SDKs if you stored top-level which flags have which bandits. Then you know if a flag involves bandits or not from the get go and can check if assignments is empty.
I do agree it's kind of a pain though.
There was a problem hiding this comment.
Yeah, I store that mapping and I can easily mirror the behavior, but I thought that this behavior is more desirable from the user perspective (they receive correct variation instead of the default one).
Split into two workspaces, so eppo_rb is in its workspace and we can commit its Cargo.lock file.
WIP: missing testing on Ruby side