createReshardingStateMachine() performs two non-atomic steps for a resharding participant: 1) ensureStateDocumentInserted() which persists the state document and 2) StateMachine::getOrCreate() — constructs/registers the in-memory PrimaryOnlyService instance that actually drives the operation and is responsible for eventually deleting its own document on completion or abort.

If step 1 succeeds but step 2 then throws (e.g. the operation is interrupted — observed in practice as _shardsvrReshardRecipientInitialize returning Interrupted immediately after the insert), the result is a persisted state document with no in-memory machine ever attached to it.

This is permanent within the current primary's term: recovery of persisted documents only happens on stepUp, and _shardsvrAbortReshardCollection's cleanup only calls abort() on an instance it can find via tryGetReshardingStateMachine — if none exists, it does nothing. The result is a permanently orphaned document, and the coordinator can never observe the participant as cleaned up after aborting.