New Relic
Alerts & Applied Intelligence
Detect issues with NRQL conditions and reduce noise with correlation and anomaly detection.
New Relic alerting is built on alert conditions, each typically an NRQL query with a threshold evaluated over a window. Conditions belong to policies, and policies feed workflows that route notifications to destinations (PagerDuty, Slack, email, webhooks). Static thresholds handle known limits; baseline conditions learn a metric's normal pattern and alert on deviation, catching problems a fixed number would miss.
A control room that not only sounds alarms but also recognises that ten alarms are really one fire, mutes the echoes, and hands the crew a single briefing.
Key Concepts
1
Applied Intelligence (AIOps) layers noise reduction on top. It correlates related incidents into a single issue so a cascade of alerts becomes one actionable notification, suppresses duplicates, and surfaces anomalies proactively. The aim mirrors good SRE practice: page on user-visible symptoms, group the rest, and give responders context (which entities, what changed) rather than a wall of separate alerts.
2
What the interviewer is probing.
What the interviewer is probing.
3
1. "How are alerts structured?" *Probing:* the model. *Stalls:* "Rules." *Moves up:* conditions
within a policy, producing incidents routed by workflows to destinations.
1. "How are alerts structured?"
4
2. "What is a baseline condition?" *Probing:* the feature. *Stalls:* "A threshold." *Moves up:*
a dynamic threshold learned from history, which suits metrics with a daily or weekly shape better
than a fixed number.
2. "What is a baseline condition?"
5
3. "What does applied intelligence add?" *Probing:* the value. *Stalls:* "AI." *Moves up:*
correlating related incidents into one issue so a single failure does not page five teams
separately.
3. "What does applied intelligence add?"
6
4. "What is the caution with anomaly detection?" *Probing:* the honest view. *Stalls:* "None."
*Moves up:* it learns whatever was normal, including a long-standing problem — and it cannot explain
itself, so it is a hint, not a verdict.
4. "What is the caution with anomaly detection?"