Abstract

A method is disclosed for detecting two silent failure modes of an object-store backup repository that conventional dashboards render as healthy. First, a write-failure inference: because a backup controller validates a storage location by reading it, a repository that has stopped accepting writes keeps reporting an available location while every write fails; the disclosed detector never writes a probe object to prove writability, and instead infers write health from the outcome of the newest real backup attributed to each write target, under a two-tier rule that separates a named store rejection from a corroborated run of unexplained failures. Second, an orphan-location garbage-collection-blocked detection: expired backup records whose declared storage-location name matches no registered location can never be reclaimed, so their bytes never leave the store. The detector deliberately refuses to remap an unresolvable location name to a default, because the unresolvable name is the condition being reported, and reports blocked reclamation as its own finding, distinct from the free-capacity finding, so the alarm names the cause and its remediation rather than only the symptom.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Share

COinS