perf: optimize free model usage IP limit counting in SQL

[kiloconnect]

Summary

• Optimizes the free model usage stats query by moving IP limit counting from JavaScript to SQL
• Replaces fetching all per-IP rows and iterating in JavaScript with a SQL subquery using HAVING COUNT(*) >= threshold
• Avoids transferring potentially large result sets to the application layer

Changes

Before:

SELECT ip_address, COUNT(*) as request_count
FROM free_model_usage
WHERE created_at >= NOW() - INTERVAL '1 hours'
GROUP BY ip_address

Then iterate in JavaScript to count IPs with request_count >= 200

After:

SELECT COUNT(*) FROM (
  SELECT ip_address
  FROM free_model_usage
  WHERE created_at >= NOW() - INTERVAL '1 hours'
  GROUP BY ip_address
  HAVING COUNT(*) >= 200
) sub

Performance Impact

• Reduces data transfer between database and application
• Eliminates JavaScript iteration overhead
• More efficient for large datasets with many IPs

File Changed

• src/app/admin/api/free-model-usage/stats/route.ts:53-91

[kiloconnect]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Clean optimization that pushes IP-at-limit counting from JavaScript to the database via SQL subqueries with HAVING COUNT(*) >= threshold. Both files apply the same pattern consistently:

• The >= comparison matches the original JS logic
• sql.raw() is only used with hardcoded numeric constants (no injection risk)
• ANONYMOUS_FILTER composes correctly inside the Drizzle sql template
• COUNT(*) from a subquery always returns exactly one row; the ?. / ?? 0 fallback is good defensive coding
• Response shape is unchanged — no breaking API changes

Files Reviewed (2 files)

• src/app/admin/api/free-model-usage/stats/route.ts - 0 issues
• src/app/admin/api/promoted-model-usage/stats/route.ts - 0 issues