You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Records port: a failed Run's outcome read once, not again every pass #285
Part of the epic "Epic: slice 1, first approval. A Release on next reviews a pull request to this repo, and the machine user's approval stands".
Why.adapters/records.py reads each completed Run's outcome from its artifact (#279, decision 0045) for every Run that did not succeed, on every pass. That is three requests a Run: the list, the redirect and the storage. The run list is never bounded by date, so the cost grows with every failed Run still listed, up to retention.
Done when
A failed Run's outcome is read once and kept by run id, outside the Kernel, which keeps nothing; or the outcomes read are bounded to the Runs an attempt window can count (max_attempts since a person with write access acted). The choice and its staleness rule are ruled in this story.
What a pass spends on outcomes is in the pass record, and a test holds the bound, with a mutant.
Activity
Records port: a failed Run's outcome read once for each attempt, kept…