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
Re-home the Wedding backup denial proof and remove cutover tooling that needs the retired shared store #4480
#3253 retires the Wedding tenant's access to the shared backup credential: the wedding-db ObjectStore and the wedding-db-backup-r2 Secret leave the wedding-app namespace. Three operator tools were written for the cutover and assume both still exist there:
Verify Wedding Backup Denial (scripts/verify-wedding-backup-denial.sh and its pod script) mounts the shared credential next to the dedicated one, as the positive control and to check that a refused write left no trace. It passed once before retirement (run 37234932425) and cannot pass again in its current form. docs/dr/runbook.md Scenario 7 still wants this proof after every rotation of the dedicated token.
Recover Wedding DB Incident (workflow and script, about 1,100 lines with tests) is pinned to the object identities of the 2026-09-09 incident and to the shared store. Its last successful run was on 2026-09-13.
All three now refuse at their first precondition. That is safe, but CI still runs their fixture suites on every pull request, and the runbook names a proof nobody can run.
Expected outcome
A dedicated-token rotation can again be followed by an observed denial, without the tenant namespace ever holding the shared credential, and the repository no longer carries one-shot tooling that cannot run.
Acceptance criteria
The denial proof runs again with the shared credential kept out of the wedding-app namespace (for example from a platform-owned namespace that already holds it, using a run-owned copy of the lower-privilege dedicated credential), keeping its positive controls and its no-trace check.
A protected main run of the re-homed proof reports the denial.
The catalogue mirror and the incident recovery tooling are removed, with their CI steps, or a recorded reason keeps them.
docs/dr/runbook.md, docs/dr/velero-cnpg.md and docs/dr/restore-drill.md describe only tooling that can run.
Closing as a duplicate of #4482. The canonical issue includes every acceptance criterion here, plus the tenant-owned database reference. Its repair is now claimed and in progress. The protected-main denial receipt remains required; duplicate closure does not mark that outcome delivered.
Activity
devantler commentedon Oct 5, 2026
Closing as a duplicate of #4482. The canonical issue includes every acceptance criterion here, plus the tenant-owned database reference. Its repair is now claimed and in progress. The protected-main denial receipt remains required; duplicate closure does not mark that outcome delivered.