fix(cost-insights): bound rollup transaction lifecycle

[jeanduplessis]

Summary

Prevent Cost Insights rollup transactions from outliving their tracked request lifecycle and creating prolonged database lock convoys.

Why this change is needed

Concurrent AI Gateway rollup writes serialize on one owner-hour unique-index entry. Detached post-commit transactions could finish SQL without reaching commit or rollback before serverless execution ended, retaining locks while queued writers consumed database connections.

How this is addressed

• Attach organization usage and Cost Insights follow-up transactions to the existing tracked usage promise while preserving separation from the billing transaction.
• Bound best-effort Cost Insights capture with transaction-local lock, statement, and idle-transaction timeouts without inline retries.
• Keep optional failures from failing canonical usage recording and report only privacy-safe PostgreSQL error metadata.
• Expand hot-key concurrency, lock-timeout, lifecycle, and canonical-repair regression coverage.

Human Verification

• Awaited post-commit lifecycle helper smoke check passed.
• Five-focus code review found no logic, data safety, security, resource, or type-contract issues.

Reviewer Notes

Human Reviewer Flags

• Cost Insights capture remains post-commit and best-effort; timed-out deltas remain repairable from canonical microdollar_usage but do not create a durable dirty-owner marker.
• Full batched derivation is intentionally deferred pending post-fix contention measurements.

Code Reviewer Agent

Code Reviewer Notes

• Production AI Gateway usage already runs in an outer Next.js after() callback. Awaiting follow-up work extends that tracked promise while remaining valid for direct admin and script callers that lack request context.
• AI Gateway Cost Insights capture uses 2s lock, 5s statement, and 10s idle-in-transaction limits. Other billing-critical rollup callers are unchanged.
• Database-backed Jest tests were not run locally because Docker was unavailable; CI should execute added regression coverage.

[kilo-code-bot]

[WARNING]: Timeout failures leave no durable repair signal

The new lock and statement timeouts can roll back captureCostInsightSpend, including its dirty-owner write, but this error path only logs the failure. Global rollup coverage therefore remains complete, so the hourly sweep can read a missing completed-hour rollup as zero instead of falling back to canonical usage, causing anomaly and threshold alerts to be missed until manual reconciliation. Record a degraded owner-hour/repair request, or otherwise force canonical fallback, before swallowing this failure.

---

Reply with @kilocode-bot fix it to have Kilo Code address this issue.

[kilo-code-bot]

Code Review Summary

Status: 1 Issues Found | Recommendation: Address before merge

Overview

Severity	Count
CRITICAL	0
WARNING	1
SUGGESTION	0

Issue Details (click to expand)

WARNING

File	Line	Issue
apps/web/src/lib/ai-gateway/processUsage.ts	549	Timeout failures leave no durable repair signal

Fix these issues in Kilo Cloud

Files Reviewed (5 files)

• apps/web/src/lib/ai-gateway/processUsage.ts - 1 issue
• apps/web/src/lib/ai-gateway/usage-post-commit-work.test.ts - 0 issues
• apps/web/src/lib/ai-gateway/usage-post-commit-work.ts - 0 issues
• apps/web/src/lib/cost-insights/rollup-maintenance.integration.test.ts - 0 issues
• packages/db/src/cost-insights-rollups.test.ts - 0 issues

---

_{Reviewed by gpt-5.6-sol-20260709 · Input: 126.9K · Output: 16.4K · Cached: 1.8M}

_{Review guidance: REVIEW.md from base branch main}