Question

Your SRE team receives dozens of pages each week for a payments API. Many pages fire on internal signals such as high CPU utilization, elevated JVM garbage-collection time, and thread-pool saturation, even though users experience no degradation during those events. Leadership wants alerting that reflects actual customer pain and reduces pager fatigue, while still catching genuine reliability problems. What is the most appropriate change to your alerting strategy?

Accepted Answer

Alert primarily on symptom-based SLIs (such as request success rate and latency against the SLO) and demote the resource-utilization metrics to dashboards for diagnosis.

Answer

Keep all existing cause-based alerts but raise their thresholds so they fire less frequently.

Answer

Route all CPU, GC, and thread-pool alerts to a lower-priority ticket queue and continue paging on every one of them during business hours.

Answer

Add more cause-based alerts covering memory, disk, and network saturation so the team can catch every possible root cause before users notice.