How can you optimize Prometheus scraping for a clustered BetterDB Monitor deployment?

Point Prometheus at the single /api/prometheus/metrics endpoint on the monitor instance itself, rather than trying to scrape each Valkey node directly – BetterDB already aggregates cluster topology, SLOT-STATS, and per-node metrics behind that one exporter, so scraping the nodes separately would duplicate work and lose that aggregation.

Keep ANOMALY_PROMETHEUS_INTERVAL_MS (default 30000ms) aligned with your Prometheus scrape_interval, so anomaly summary values aren't sampled mid-update or lagging behind what Prometheus expects.

If you're running the Helm chart, target the betterdb-monitor Kubernetes Service in your scrape config instead of individual pod IPs, so scraping keeps working across pod restarts and rescheduling rather than breaking every time a pod's IP changes.

It's also worth setting your Prometheus scrape_timeout with some headroom above normal response time on a busy cluster: a large SLOT-STATS heatmap computation across many nodes can take noticeably longer to summarize than a typical single-instance metrics scrape, and a too-tight timeout would show up as intermittent, confusing scrape failures rather than a real outage.