Improve SQL Databases

Goal

Correct the selected SQL query, transaction, migration or connection behavior for the supported database and deployment.

Domain rules

• Fix ownership at database constraint/transaction/query boundary before adding application-side coordination.
• Keep transactions appropriately bounded and avoid unrelated network work inside them. Retry a complete safe operation only for classified transient failures; reconcile ambiguous commits instead of blindly repeating effects.
• Use online/staged migration patterns supported by the target database and make backfills resumable with progress/checkpoint and load control.
• Set pool and timeout values from deployment-wide budgets and expose acquire wait, usage, errors, query latency, locks, and lag.
• Preserve read/write routing and tenant enforcement during failover and mixed-version rollout; regenerate clients only from authoritative schema changes.

Validation

• Run repository SQL/migration tests against the supported database version, not only mocks or an incompatible in-memory engine.
• Exercise concurrent invariant conflicts, pool saturation, cancellation, lock timeout/deadlock, migration restart, replica lag/read-after-write, and reconnect/failover where feasible.
• Capture plans and bounded load measurements with representative data shape; state production-volume/topology assumptions explicitly.

Correct the selected constraint/transaction, query/index, migration or pool owner using the same records, role and workload as the failure. Preserve supported old/new readers and rollback/forward-fix conditions. Feed the candidate and controlled interleaving into the load/lock check. Recheck failed values, cleanup or compatibility before Patch Checking, without masking the problem with a larger global pool or timeout.