billing-health

Billing Health

Monitor the OpenMeter metering pipeline using Supabase MCP for direct SQL queries against the outbox ledger.

Prerequisites

• Supabase MCP: execute_sql tool
• Database access to the openmeter_event_ledger table

Monitoring Procedures

1. Outbox Health Check

Use Supabase MCP execute_sql with this query to assess queue health:

SELECT
  COUNT(*) FILTER (WHERE sent_at IS NULL AND attempts < 10) AS pending,
  COUNT(*) FILTER (WHERE sent_at IS NOT NULL) AS sent,
  COUNT(*) FILTER (WHERE attempts >= 10) AS dead_letter,
  COUNT(*) FILTER (WHERE lease_until IS NOT NULL AND lease_until > now()) AS leased,
  COUNT(*) AS total
FROM openmeter_event_ledger
WHERE created_at > now() - interval '24 hours'

See references/outbox-queries.md for all outbox queries.

Interpret results using references/thresholds.md:

• pending > 500 → HIGH QUEUE DEPTH
• dead_letter > 0 → DEAD LETTERS — events failed 10+ times
• Stuck leases (expired but unsent) → outbox worker may be down
• sent = 0 with total > 0 → NO EVENTS DELIVERED — check API connectivity

2. Per-Org Usage Breakdown

See references/usage-queries.md for SQL queries that break down usage by org, provider, and model.

3. Dead Letter Recovery

When dead letters exist, retry them with this procedure:

1. Check the top errors first (see outbox-queries.md "Top Errors" query)
2. Fix the root cause (usually API key, timeout, or connectivity)
3. Run the retry query (see outbox-queries.md "Dead Letter Retry" query)
4. Re-check outbox health to verify events are being delivered

4. Usage Anomaly Detection

See references/anomaly-detection.md for the spike/drop detection algorithm.