fix(server): serialize sandbox restart authentication - #3485
Conversation
48e2473 to
6215d97
Compare
b403498 to
bea0fdf
Compare
Signed-off-by: Drew Newberry <anewberry@nvidia.com>
6215d97 to
8353f52
Compare
bea0fdf to
924f043
Compare
|
Label |
PR Review StatusThe initial code review found no validated blocking findings. The sandbox lifecycle change requires the standard E2E suite, but applying Action required: A maintainer must open Branch E2E run 35482747929 and choose Re-run all jobs so the suite executes with Blocking findings:
Carried findings:
Non-blocking suggestions:
Gator metadata
|
elezar
left a comment
There was a problem hiding this comment.
Reviewed the lifecycle and authentication changes. Moving restart identity preparation under lifecycle serialization and committing the identity with the Starting transition in one CAS closes the stale-credential window; idempotent retries correctly reuse the persisted winning identity. All required checks, including Core E2E, are green. Approved.
Monitoring CompleteMonitoring is complete because this PR has merged. Final status: the current head passed the required branch, Helm, Trivy, and Core E2E checks, received maintainer approval, and reached I removed the active |
Summary
A stopped sandbox could rotate its runtime identity before acquiring the compute lifecycle locks, allowing reconciliation or another start to invalidate the CAS or launch with credentials that no longer matched persisted state. Commit the identity rotation and
Startingtransition in one resource-version update, retry conflicts, and make idempotent starts reuse the winning identity.Related Issue
No issue required: localized concurrency bug confirmed across the reviewed lifecycle failures and requested directly for implementation.
Changes
Startingphase.Testing
mise run pre-commitpassesmise run testpassesChecklist