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
Exercise the fenced provider-close branch and operator recovery in E2E again #271
Follow-up to #264 (bounded shutdown grace) and #251 (operator recovery).
Problem
With the shutdown grace, Close during a provider session now lets the guest return auth and write its stopped receipt, so the E2E close-provider case takes the "Close proves completion" branch (measured on mini at 89c36a1: 58 checks, job ends incomplete_expired, one INCOMPLETE status, both auth stores usable). The fenced branch — guest does not return within the grace, auth fenced, root retained — is now reached only if the guest hangs, which no E2E scenario produces. As a result close-provider-reopen and close-provider-recover no longer run on a host, and #251's operator recovery keeps only unit and fault coverage.
Acceptance criteria
An E2E scenario deterministically reaches the fenced branch with real VMs, for example a fake-model/fixture mode that keeps the guest from returning auth within the grace (without weakening any production bound or adding a production hook that changes behaviour).
close-provider-reopen and close-provider-recover run from that root again and pass every existing check, including the recovery receipt, retired → usable auth sequence and PR 2's review.
The normal close-provider grace branch keeps running in the default case set.
Validation
All E2E cases on mini from one clean commit, one VM at a time behind the memory gate.
Activity
Require the grace branch in close-provider and name close-provider-fe…
Reach the fenced provider-close branch in E2E again with close-provid…