# ExportCenter Alert Rules
# SLO Burn-rate alerts for export service reliability

groups:
  - name: export-center-slo
    interval: 30s
    rules:
      # SLO: 99.5% success rate target
      # Error budget: 0.5% (432 errors per day at 86400 requests/day)

      # Fast burn - 2% budget consumption in 1 hour (critical)
      - alert: ExportCenterHighErrorBurnRate
        expr: |
          (
            sum(rate(export_runs_failed_total[1h]))
            /
            sum(rate(export_runs_total[1h]))
          ) > (14.4 * 0.005)
        for: 2m
        labels:
          severity: critical
          service: export-center
          slo: availability
        annotations:
          summary: "ExportCenter high error burn rate"
          description: "Error rate is {{ $value | humanizePercentage }} over the last hour, consuming error budget at 14.4x the sustainable rate."
          runbook_url: "https://docs.stellaops.io/runbooks/export-center/high-error-rate"

      # Slow burn - 10% budget consumption in 6 hours (warning)
      - alert: ExportCenterElevatedErrorBurnRate
        expr: |
          (
            sum(rate(export_runs_failed_total[6h]))
            /
            sum(rate(export_runs_total[6h]))
          ) > (6 * 0.005)
        for: 5m
        labels:
          severity: warning
          service: export-center
          slo: availability
        annotations:
          summary: "ExportCenter elevated error burn rate"
          description: "Error rate is {{ $value | humanizePercentage }} over the last 6 hours, consuming error budget at 6x the sustainable rate."
          runbook_url: "https://docs.stellaops.io/runbooks/export-center/elevated-error-rate"

  - name: export-center-latency
    interval: 30s
    rules:
      # SLO: 95% of exports complete within 120s
      # Fast burn - p95 latency exceeding threshold
      - alert: ExportCenterHighLatency
        expr: |
          histogram_quantile(0.95,
            sum(rate(export_run_duration_seconds_bucket[5m])) by (le)
          ) > 120
        for: 5m
        labels:
          severity: warning
          service: export-center
          slo: latency
        annotations:
          summary: "ExportCenter high latency"
          description: "95th percentile export duration is {{ $value | humanizeDuration }}, exceeding 120s SLO target."
          runbook_url: "https://docs.stellaops.io/runbooks/export-center/high-latency"

      # Critical latency - p99 exceeding 5 minutes
      - alert: ExportCenterCriticalLatency
        expr: |
          histogram_quantile(0.99,
            sum(rate(export_run_duration_seconds_bucket[5m])) by (le)
          ) > 300
        for: 2m
        labels:
          severity: critical
          service: export-center
          slo: latency
        annotations:
          summary: "ExportCenter critical latency"
          description: "99th percentile export duration is {{ $value | humanizeDuration }}, indicating severe performance degradation."
          runbook_url: "https://docs.stellaops.io/runbooks/export-center/critical-latency"

  - name: export-center-capacity
    interval: 60s
    rules:
      # Queue buildup warning
      - alert: ExportCenterHighConcurrency
        expr: sum(export_runs_in_progress) > 50
        for: 5m
        labels:
          severity: warning
          service: export-center
        annotations:
          summary: "ExportCenter high concurrency"
          description: "{{ $value }} exports currently in progress. Consider scaling or investigating slow exports."
          runbook_url: "https://docs.stellaops.io/runbooks/export-center/high-concurrency"

      # Stuck exports - exports running longer than 30 minutes
      - alert: ExportCenterStuckExports
        expr: |
          histogram_quantile(0.99,
            sum(rate(export_run_duration_seconds_bucket{status!="completed"}[1h])) by (le)
          ) > 1800
        for: 10m
        labels:
          severity: warning
          service: export-center
        annotations:
          summary: "ExportCenter potentially stuck exports"
          description: "Some exports may be stuck - 99th percentile duration for incomplete exports exceeds 30 minutes."
          runbook_url: "https://docs.stellaops.io/runbooks/export-center/stuck-exports"

  - name: export-center-errors
    interval: 30s
    rules:
      # Specific error code spike
      - alert: ExportCenterErrorCodeSpike
        expr: |
          sum by (error_code) (
            rate(export_runs_failed_total[5m])
          ) > 0.1
        for: 5m
        labels:
          severity: warning
          service: export-center
        annotations:
          summary: "ExportCenter error code spike: {{ $labels.error_code }}"
          description: "Error code {{ $labels.error_code }} is occurring at {{ $value | humanize }}/s rate."
          runbook_url: "https://docs.stellaops.io/runbooks/export-center/error-codes"

      # No successful exports in 15 minutes (when there is traffic)
      - alert: ExportCenterNoSuccessfulExports
        expr: |
          (
            sum(rate(export_runs_total[15m])) > 0
          )
          and
          (
            sum(rate(export_runs_success_total[15m])) == 0
          )
        for: 10m
        labels:
          severity: critical
          service: export-center
        annotations:
          summary: "ExportCenter no successful exports"
          description: "No exports have completed successfully in the last 15 minutes despite ongoing attempts."
          runbook_url: "https://docs.stellaops.io/runbooks/export-center/no-successful-exports"

  - name: export-center-deprecation
    interval: 5m
    rules:
      # Deprecated endpoint usage
      - alert: ExportCenterDeprecatedEndpointUsage
        expr: |
          sum(rate(export_center_deprecated_endpoint_access_total[1h])) > 0
        for: 1h
        labels:
          severity: info
          service: export-center
        annotations:
          summary: "Deprecated export endpoints still in use"
          description: "Legacy /exports endpoints are still being accessed at {{ $value | humanize }}/s. Migration to v1 API recommended."
          runbook_url: "https://docs.stellaops.io/api/export-center/migration"