Skip to main content
Alerts turn passive metrics into action. You define rules on pipeline outcomes, duration, DLQ depth, freshness, and anomaly signals, then route notifications to email, chat, PagerDuty-style endpoints, or SMS where your plan supports it.

​
Alert rules

Create a rule from a template or from scratch:
  1. Scope – pipeline, tag, connection, or workspace-wide policies
  2. Condition – failure, success after prior failure, duration > threshold, row count delta beyond band
  3. Window – rolling 1h/24h/7d evaluation windows
  4. Severity – informational vs paging
  5. Channels – map to notification preferences
Start with failure + duration alerts; add volume and anomaly alerts once baselines stabilize.

​
Notification channels

Best for non-urgent summaries; include run links in the template so recipients land in context.

​
Smart anomaly alerts

Anomaly alerts use forecasting on volume, latency, or error rates. They catch issues that static thresholds miss—like a job that always succeeds but returns zero rows on a day you expect millions.
Allow a few weeks of history before enabling strict anomaly paging; early models misread seasonality.
Read the anomaly card’s drivers (which metric component spiked) before muting; muting hides real outages.
Alert storms erode trust. Use correlation suppression so the same root cause does not open fifty tickets.

Dashboard

Spot trends before they breach alert thresholds.

Diagnostics

Investigate anomalies surfaced by smart alerts.