S3 Health States

KafScale brokers publish a live S3 health state based on latency and error-rate sampling windows.

State definitions

• HEALTHY: S3 latency and error rate below warning thresholds.
• DEGRADED: Latency or error rate above warning thresholds.
• UNAVAILABLE: Latency or error rate above critical thresholds.

Metrics and thresholds

Metrics:

• kafscale_s3_health_state (label state)
• kafscale_s3_latency_ms_avg
• kafscale_s3_error_rate
• kafscale_s3_state_duration_seconds

Thresholds (configurable):

• KAFSCALE_S3_LATENCY_WARN_MS
• KAFSCALE_S3_LATENCY_CRIT_MS
• KAFSCALE_S3_ERROR_RATE_WARN
• KAFSCALE_S3_ERROR_RATE_CRIT
• KAFSCALE_S3_HEALTH_WINDOW_SEC

Alerting integration

Wire S3 health transitions into Prometheus rules or your alerting stack. A common pattern:

• Alert on state="unavailable" immediately.
• Alert on state="degraded" if sustained for more than a few minutes.
• Track latency/error trends to tune thresholds per region.