groups:
- name: alert.rules
  rules:
  - alert: TiFlash_tmt_merge_duration
    expr: histogram_quantile(0.99, sum(rate(tiflash_tmt_merge_duration_seconds_bucket[1m])) BY (le, instance)) > 600
    for: 1m
    labels:
      env: ENV_LABELS_ENV
      level: emergency
      expr:  histogram_quantile(0.99, sum(rate(tiflash_tmt_merge_duration_seconds_bucket[1m])) BY (le, instance)) > 600
    annotations:
      description: 'cluster: ENV_LABELS_ENV, instance: {{ $labels.instance }}, values:{{ $value }}'
      value: '{{ $value }}'
      summary: TiFlash tmt merge duration 99th percentile is above 600s

  - alert: TiFlash_tmt_write_parts_duration
    expr: histogram_quantile(0.99, sum(rate(tiflash_tmt_write_parts_duration_seconds_bucket[1m])) BY (le, instance)) > 8
    for: 1m
    labels:
      env: ENV_LABELS_ENV
      level: emergency
      expr:  histogram_quantile(0.99, sum(rate(tiflash_tmt_write_parts_duration_seconds_bucket[1m])) BY (le, instance)) > 8
    annotations:
      description: 'cluster: ENV_LABELS_ENV, instance: {{ $labels.instance }}, values:{{ $value }}'
      value: '{{ $value }}'
      summary: TiFlash tmt write parts duration 99th percentile is above 8s

  - alert: TiFlash_schema_error
    expr: increase(tiflash_schema_apply_count{type="failed"}[15m]) > 0
    for: 1m
    labels:
      env: ENV_LABELS_ENV
      level: emergency
      expr:  increase(tiflash_schema_apply_count{type="failed"}[15m]) > 0
    annotations:
      description: 'cluster: ENV_LABELS_ENV, instance: {{ $labels.instance }}, values:{{ $value }}'
      value: '{{ $value }}'
      summary: TiFlash schema error

  - alert: TiFlash_schema_apply_duration
    expr: histogram_quantile(0.99, sum(rate(tiflash_schema_apply_duration_seconds_bucket[1m])) BY (le, instance)) > 20
    for: 1m
    labels:
      env: ENV_LABELS_ENV
      level: emergency
      expr:  histogram_quantile(0.99, sum(rate(tiflash_schema_apply_duration_seconds_bucket[1m])) BY (le, instance)) > 20
    annotations:
      description: 'cluster: ENV_LABELS_ENV, instance: {{ $labels.instance }}, values:{{ $value }}'
      value: '{{ $value }}'
      summary: TiFlash schema apply duration 99th percentile is above 20s

  - alert: TiFlash_raft_read_index_duration
    expr: histogram_quantile(0.99, sum(rate(tiflash_raft_read_index_duration_seconds_bucket[1m])) BY (le, instance)) > 3
    for: 1m
    labels:
      env: ENV_LABELS_ENV
      level: critical
      expr:  histogram_quantile(0.99, sum(rate(tiflash_raft_read_index_duration_seconds_bucket[1m])) BY (le, instance)) > 3
    annotations:
      description: 'cluster: ENV_LABELS_ENV, instance: {{ $labels.instance }}, values:{{ $value }}'
      value: '{{ $value }}'
      summary: TiFlash raft read index duration 99th percentile is above 3s

  - alert: TiFlash_raft_wait_index_duration
    expr: histogram_quantile(0.99, sum(rate(tiflash_raft_wait_index_duration_seconds_bucket[1m])) BY (le, instance)) > 2
    for: 1m
    labels:
      env: ENV_LABELS_ENV
      level: critical
      expr:  histogram_quantile(0.99, sum(rate(tiflash_raft_wait_index_duration_seconds_bucket[1m])) BY (le, instance)) > 2
    annotations:
      description: 'cluster: ENV_LABELS_ENV, instance: {{ $labels.instance }}, values:{{ $value }}'
      value: '{{ $value }}'
      summary: TiFlash raft wait index duration 99th percentile is above 2s
