Connect a Metrics source
Add the source address, authentication, and connection settings. Check availability before defining metric templates. For Prometheus-compatible queries, confirm the selected series belong to the intended workload.
Define metric templates
Templates hold queries, variables, units, and aggregation. Modules reference them and render queries for their targets. Start with a built-in preset and adapt it to real cluster labels.
- Align CPU, memory, and request scope; distinguish workload totals from per-Pod use.
- Requests and limits are configuration; actual use is an observation.
- OOM, JVM heap pressure, and throttling provide additional protection signals.
- For empty queries, inspect labels, filters, the source, and time range.
Datasets and historical quality
Datasets align sample intervals, aggregation, conversion, precision, and gap handling. Inspect coverage, gaps, and freshness against model sample and season requirements.
Handle missing, invalid, and stale samples through explicit rules. Do not interpret gaps as falling demand. Recommendations should explain input quality and degradation.
Units and time
Specify conversions between cores and millicores, bytes and MiB, and counts and rates. Storage uses UTC; pages display UTC+8. Schedules use their explicitly configured IANA timezone.