- Shipped
- October 21, 2026 at 6:11 PM UTC
- Author
- Kamo
- Commit
- 7ca04ac
The pipeline's two reads were N+1 on the round trip: - GET /mortgage-apps loaded plain AppMortgage rows, and each one then fired a SELECT per inverse @OneToOne (loanFile, loan, property, loanOriginator). Hibernate cannot proxy or batch the mappedBy side, so default_batch_fetch_size never helped: four round trips per application. It now fetch-joins them. - GET /mortgage-apps/transactions ran `select lf from LoanFile` once PER deal, and every LoanFile it hydrated eagerly dragged in its AppMortgage (non-lazy owning one-to-one) and that app's four one-to-ones. The rollup is now one scalar projection for every deal at once; no entity is hydrated. The transaction summary also carries `terminal` (completed/cancelled, the pipeline's "past" view), the assigned loan officer and the date the deal was opened, and two deal-management endpoints join the existing LO assignment: - PUT **************** rename (LOS write right) - POST **************** add a blank loan to the deal (take-application right); the loan inherits the deal's account The HQL is parsed against the shared-library mapping on this service's Hibernate in a test, since nothing else checks string-built queries before production.