Workspace Lifecycle Service (Platform)
What This Service Is For
Platform coordinates workspace construction, controlled initial loading, reset and deliberate removal. It retains the decisions and progress staff need to inspect and recover those operations. Its availability implementation evaluates lifecycle and business standing for gated workspace requests.
Boundary
Workspace administration retains ordinary customer and workspace records. Shared jobs own execution and reconciliation; Security owns authority; the Data Job Engine owns imports; export coordination owns packaging and delivery. Application lifecycle functions implement their participant work. Provider sending and identity concerns retain their existing service owners.
Operations in Words
Staff can request construction, inspect its attempt, retry failed construction and initialize an existing participant. Import authority opens loading, submits reset and reads reset progress; separate activation authority ends loading.
Deletion-request authority creates a decision and recovers stopped watcher or purge work. Hold authority pauses and releases reversible decisions; cancellation authority withdraws them. Deletion-reading authority exposes retained requests and their notices. Immediate removal is refused. The exact route permission and capability mappings remain in the structured operation rows.
Events and Jobs
Construction attempts publish keyed HQ outcome notifications and attempt requester messages. The reconciliation hook repairs stranded construction; it is an in-process callback rather than a business event. Four state machines perform construction, watching, purge and reset. Operator scripts inspect leftovers and disabled candidates or reapply ledger-recorded removal after restore. Notice failures feed an operations alarm. The reviewed lifecycle interface performs no model inference.
Consistency Guarantees
Normal admission and irreversible handoff commit their domain records with new job records before execution starts. Provisioning retry and deletion recovery have separate-commit gaps. Participant commits also precede result accounting, so repeated removal can converge while counts remain incomplete. Availability facts expire per process; HQ record routes outside the gate need their own safeguards.
Purge finalization is not one transaction: failure after workspace removal can prevent completion and ordinary retry. Completion also precedes the external ledger write, whose exhausted failure has no ordinary recovery route. Notice records prevent duplicate obligations but do not ensure one message or eventual delivery.