Metrics

Metric alerts carry:

• alertID, alertName, alertValue — threshold and value
• group, groupValue — the dimension (if grouped)
• query — the metric query
• thresholdWindow — hold duration
• metricsLink — deep link
• timeRange — alert window

Investigation shape

1. Replay the alert’s metric query using the query-aggregations tool, same product_type, query, and time window as the alert. Confirm you see the same value that breached.
2. Expand the time window just enough to see onset — if the threshold was crossed at T, pull query-aggregations from T - 2 * thresholdWindow to T to see the rise.
3. Drill into the underlying records. A metric alert on errors points to specific error groups; on sessions to specific sessions; on traces to specific traces. Pull the relevant record-level data for the top contributing dimension.
4. Check external triggers. Deploy correlation, flag flip, traffic surge, dependency issue.

What goes in the diagnosis

• What triggered — alert name, threshold, alertValue, and (if grouped) which dimension value crossed it.
• Likely cause — specific underlying records (error group IDs, trace IDs, session IDs) that drove the metric up. Quote one or two.
• Scope — affected count, time-window pattern, whether this is a sharp spike or a drift.
• Next steps — concrete actions per the root cause found; if the metric reflects underlying behavior that needs fixing, point to the record-level remediation.

Common mistakes

• Only reporting the metric value without drilling into record-level evidence. The metric is a summary; the cause lives in the records.
• Mis-attributing a spike to a specific cause without correlating with a deploy/flag/dependency timeline.
• Querying too-wide a time range and washing out the signal. Stay near the alert window.