Skip to main content

Operations, outages and recovery

Recovery should address the observed boundary. Do not replace an entire profile, credential store or node identity merely because a component cannot finish an operation.

SymptomDistinguish firstPreserve
Tool returned promptly but no target proof yetAdmission versus deliveryOriginal operation/body/target binding
Tool or SSH timed outUncertainty versus confirmed rejectionNo automatic resend
Recipient absent/staleCorrect surface, account, native identity and registry sourceExisting provider history and admitted requests
Provider tools missingInstallation, extension activation and provider permissionsOwned configuration and unrelated entries
Account signed in but peers absentDevice-connect consent, Relay enablement and approvalExisting host keys and independent pairing
Account authority unavailableBackoff and account-only freshness expiryRefuse new stale account work; preserve independent grants
Sign-out incompleteLocal usable session versus server-family revocationVisible recovery record
Update refusedSignature/protocol/ownership/shutdown/migration boundaryVerified active version and transaction material
Installation warningsWhich explicit selections remain incompleteHealthy independent installations

Outage behavior​

Reconnection is a transport action. It must not replay a logical provider Send. Account hints may be delayed, so reconnect/fallback obtains authority independently. The five-minute source lease bounds new account-only admission even when requests fail.

A provider login outage is not an AxiOwl license failure. A2A endpoint completion and remote SSH command completion are also separate from native provider delivery.

Backup and restore​

Back up exact application/runtime state with its supported ownership and identity. Include the records needed for deduplication and recovery. Protect message bodies, trust material and credentials in backups.

Do not restore only a database while discarding its required identity/sidecar/transaction relationship. A copied path does not establish correct owner or native host identity.

No universal disaster-recovery procedure or tested recovery-time objective is claimed here. Follow the matching component's published migration/recovery contract.

Escalation evidence​

Report the exact operation, OS/user/surface, versions, observed error and whether any part may already have been accepted. Preserve unknown times instead of inventing them.

A suggested investigation is not authorization for a provider test, reinstall, reboot, credential reset or live fix. During a restricted task, use source/local evidence and ask only for information or action that is actually required.

See observability, installation, credentials and source status.