Notion Observability
Overview
Instrument Notion API calls with metrics, structured logging, and alerting. Track request rates, latencies, error rates, and rate limit headroom across a full observability stack: an instrumented client wrapper, Prometheus metrics, structured logging via pino, health check endpoints, and alerting rules.
Prerequisites
@notionhq/client v2+ installed (npm install @notionhq/client)- Python alternative:
notion-client (pip install notion-client)
- Prometheus-compatible metrics backend (optional: Grafana, Datadog, or CloudWatch)
- Structured logging library:
pino (Node.js) or structlog (Python)
Authentication
All snippets read the integration token from the NOTION_TOKEN environment variable (never hard-code it) and pass it as auth to the Notion client constructor. Store it in your secret manager and inject it at runtime. The health check and metrics endpoints below expose no secrets — only aggregate counters and status.
Instructions
The workflow layers three pieces. Build them in order; each is self-contained. Full code for every step lives in references/implementation.md.
Step 1: Instrumented client wrapper
Wrap every Notion call so timing, error classification, rate-limit detection, and structured logging happen automatically. The wrapper accumulates per-operation latency buckets and exposes getMetrics() for avg/p95. Skeleton:
class InstrumentedNotionClient {
async call<T>(operation: string, fn: (c: Client) => Promise<T>): Promise<T> {
const start = performance.now();
try {
const result = await fn(this.client);
this.recordLatency(operation, Math.round(performance.now() - start));
return result;
} catch (error) {
if (isNotionClientError(error) && error.code === APIErrorCode.RateLimited) {
this.metrics.rateLimitCount++;
}
throw error;
}
}
}
A Python (notion-client) equivalent is in the reference. Full TypeScript + Python wrappers: references/implementation.md (Step 1).
Step 2: Prometheus metrics export
Register a request counter, a latency histogram (buckets tuned for Notion's typical 200-800ms responses), an error counter keyed by error code, and a rate-limit gauge. Wrap calls with startTimer() and expose a /metrics endpoint for scraping. Full export code: references/implementation.md (Step 2).
Step 3: Health check, structured logging, and alerting
Add a /health/notion endpoint that probes users.me and returns 200/503 with aggregate metrics, wire pino for JSON logs with slow-query warnings (>2s), and ship Prometheus alerting rules for error-rate spikes, rate-limit exhaustion, high P95 latency, and outages. Full endpoint, logger, and alert rules: references/implementation.md (Step 3).
Output
- Instrumented Notion client tracking all API calls with per-operation latency buckets
- Prometheus metrics for request rate, latency histograms, and error counters
- Structured JSON logging via pino with slow-query warnings (>2s)
- Health check endpoint with Notion connectivity status and aggregate metrics
- Alerting rules for error rate spikes, rate limiting, high latency, and outages
Error Handling
| Issue | Cause | Solution |
|---|
| High cardinality metrics | Too many unique label values | Use fixed operation names (databases.query, pages.create) |
| Alert storms on Notion outage | All alerts fire simultaneously | Add group_wait: 30s in alertmanager config |
| Missing metrics for some calls | Not all API calls use wrapper | Enforce wrapper at architecture level |
| Log volume too high in prod | DEBUG level enabled | Set LOG_LEVEL=info or warn in production |
| P95 latency unreliable | Too few samples | Ensure minimum 100 requests in window |
| Rate limit counter never fires | Wrong error code check | Use APIErrorCode.RateLimited constant |
Examples
Ready-to-run PromQL dashboard queries and a no-Prometheus inline console metrics snippet live in references/examples.md. Quickest smoke test — error percentage in PromQL:
100 * rate(notion_errors_total[5m]) / rate(notion_requests_total[5m])
Resources
Next Steps
For incident response procedures when monitoring detects failures, see the notion-incident-runbook skill. Once metrics flow, build a Grafana dashboard from the PromQL queries in references/examples.md and tune alert thresholds to your traffic baseline.
