Monitoring
An operations practice of predefining metrics and setting threshold alerts, answering the question of whether a system is functioning normally, and distinguished from observability in its ability to locate problems.
Timeline (last 90 days)
Conventional latency and error-rate metrics are insufficient to detect content leakage.
Monitoring means predefining the metrics to watch and setting threshold alerts, and answers "is the system healthy."
Monitoring logs are written only after a call has happened, so monitoring should sit alongside per-call authorization rather than replace it.
The minimum monitoring checklist for an unattended agent is: one heartbeat per scheduled trigger, a queue-depth alert, a dead-letter queue alert, a worker error-rate alert, and a "no successful run in X hours" alert.
Monitoring can only tell you whether an agent ran and how long it ran, but cannot tell you whether the work output is intact.
Monitoring focuses on detecting known issues, while observability focuses on exploring unknown issues.
All Facts (6)
Conventional latency and error-rate metrics are insufficient to detect content leakage.
50%UnverifiedMonitoring means predefining the metrics to watch and setting threshold alerts, and answers "is the system healthy."
50%UnverifiedMonitoring logs are written only after a call has happened, so monitoring should sit alongside per-call authorization rather than replace it.
50%UnverifiedThe minimum monitoring checklist for an unattended agent is: one heartbeat per scheduled trigger, a queue-depth alert, a dead-letter queue alert, a worker error-rate alert, and a "no successful run in X hours" alert.
50%UnverifiedMonitoring can only tell you whether an agent ran and how long it ran, but cannot tell you whether the work output is intact.
50%UnverifiedMonitoring focuses on detecting known issues, while observability focuses on exploring unknown issues.
50%