Add DB pool observability, OTel pg instrumentation, and statement_timeout

[iscekic]

Summary

Adds application-side database observability and enforces the statement_timeout guard that was validated but never applied.

1. Pool metrics → Axiom (via Vercel log drain)

Logs pool.totalCount, pool.idleCount, and pool.waitingCount every 30s as structured JSON for both primary and replica pools. This gives the client-side view of connection usage per Vercel instance/region, complementing the server-side view from the new Supabase → Grafana integration.

Example log line:

{"type":"pool_metrics","region":"iad1","primary":{"total":8,"idle":5,"waiting":0},"replica":{"total":3,"idle":2,"waiting":0}}

Queryable in Axiom with:

['vercel'] | where ['fields.type'] == 'pool_metrics'

2. statement_timeout enforcement

Applies POSTGRES_MAX_QUERY_TIME (already validated as a required env var) as statement_timeout on both primary and replica pools. Previously the env var was checked at startup but never actually used — queries had no server-side timeout guard.

Not included: OTel pg instrumentation

Sentry's postgresIntegration (enabled by default) already uses @opentelemetry/instrumentation-pg under the hood, so per-query spans are already being captured in Sentry traces without any code change needed.

Context

Follow-up to #497 (connection exhaustion fix). The Supabase dashboards provide limited visibility, so we're adding application-side observability to correlate with the new Grafana dashboards.

Changes

• src/lib/drizzle.ts — pool metrics logging + statement_timeout on both pools

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Clean, well-structured change that adds two improvements to the database connection layer:

1. Statement timeout (statement_timeout) applied to both primary and replica pools using the existing POSTGRES_MAX_QUERY_TIME env var. This prevents runaway queries from holding connections indefinitely — good defensive measure.

2. Pool observability via periodic JSON-structured metrics logging (pool_metrics). Properly guarded with NODE_ENV !== 'test', uses .unref() to avoid keeping the process alive, and includes region context for log correlation.

Files Reviewed (1 file)

• src/lib/drizzle.ts - 0 issues