groups:
  - name: pg36-shop-sli-recording
    interval: 1m
    type: prometheus
    rules:
      - record: pg36_shop:sli_availability_bad_ratio:rate5m
        expr: |
          sum by (service, operation_class, environment) (
            rate(pg36_shop_request_outcomes_total{eligible="true",outcome!="good"}[5m])
          )
          /
          sum by (service, operation_class, environment) (
            rate(pg36_shop_request_outcomes_total{eligible="true"}[5m])
          )
      - record: pg36_shop:sli_availability_bad_ratio:rate30m
        expr: |
          sum by (service, operation_class, environment) (
            rate(pg36_shop_request_outcomes_total{eligible="true",outcome!="good"}[30m])
          )
          /
          sum by (service, operation_class, environment) (
            rate(pg36_shop_request_outcomes_total{eligible="true"}[30m])
          )
      - record: pg36_shop:sli_availability_bad_ratio:rate1h
        expr: |
          sum by (service, operation_class, environment) (
            rate(pg36_shop_request_outcomes_total{eligible="true",outcome!="good"}[1h])
          )
          /
          sum by (service, operation_class, environment) (
            rate(pg36_shop_request_outcomes_total{eligible="true"}[1h])
          )
      - record: pg36_shop:sli_availability_bad_ratio:rate6h
        expr: |
          sum by (service, operation_class, environment) (
            rate(pg36_shop_request_outcomes_total{eligible="true",outcome!="good"}[6h])
          )
          /
          sum by (service, operation_class, environment) (
            rate(pg36_shop_request_outcomes_total{eligible="true"}[6h])
          )
      - record: pg36_shop:sli_availability_bad_ratio:rate3d
        expr: |
          sum by (service, operation_class, environment) (
            rate(pg36_shop_request_outcomes_total{eligible="true",outcome!="good"}[3d])
          )
          /
          sum by (service, operation_class, environment) (
            rate(pg36_shop_request_outcomes_total{eligible="true"}[3d])
          )
      - record: pg36_shop:sli_latency_bad_ratio:rate5m
        expr: |
          1 -
          (
            sum by (service, operation_class, environment) (
              rate(pg36_shop_request_duration_seconds_bucket{eligible="true",le="0.25"}[5m])
            )
            /
            sum by (service, operation_class, environment) (
              rate(pg36_shop_request_duration_seconds_count{eligible="true"}[5m])
            )
          )
      - record: pg36_shop:sli_latency_bad_ratio:rate1h
        expr: |
          1 -
          (
            sum by (service, operation_class, environment) (
              rate(pg36_shop_request_duration_seconds_bucket{eligible="true",le="0.25"}[1h])
            )
            /
            sum by (service, operation_class, environment) (
              rate(pg36_shop_request_duration_seconds_count{eligible="true"}[1h])
            )
          )
      - record: pg36_shop:sli_freshness_bad_ratio:rate5m
        expr: |
          sum by (service, read_path, environment) (
            rate(pg36_shop_commit_visibility_probes_total{eligible="true",within_bound="false"}[5m])
          )
          /
          sum by (service, read_path, environment) (
            rate(pg36_shop_commit_visibility_probes_total{eligible="true"}[5m])
          )
      - record: pg36_shop:sli_freshness_bad_ratio:rate1h
        expr: |
          sum by (service, read_path, environment) (
            rate(pg36_shop_commit_visibility_probes_total{eligible="true",within_bound="false"}[1h])
          )
          /
          sum by (service, read_path, environment) (
            rate(pg36_shop_commit_visibility_probes_total{eligible="true"}[1h])
          )

  - name: pg36-postgresql-diagnostic-recording
    interval: 1m
    type: prometheus
    rules:
      - record: pg36:replica_replay_distance_bytes:max
        expr: max by (cls, ins, ip) (pg_repl_replay_diff)
      - record: pg36:archive_failures:increase15m
        expr: sum by (cls, ins, ip) (increase(pg_archiver_failed_count[15m]))
      - record: pg36:longest_transaction_seconds:max
        expr: |
          max by (cls, ins, ip, datname, state) (
            pg_activity_max_tx_duration{state=~"active|idle in transaction"}
          )
      - record: pg36:table_freeze_age:max
        expr: max by (cls, ins, ip) (pg_table_age)
      - record: pg36:dead_tuples:sum
        expr: sum by (cls, ins, ip, datname) (pg_table_n_dead_tup)
      - record: pg36:exporter_unavailable:max
        expr: max by (cls, ins, ip) (1 - pg_exporter_up)

  - name: pg36-observation-meta-recording
    interval: 1m
    type: prometheus
    rules:
      - record: pg36:vmalert_rule_errors:sum
        expr: |
          sum(vmalert_alerting_rules_errors_total)
          +
          sum(vmalert_recording_rules_errors_total)
      - record: pg36:vmalert_missed_iterations:increase15m
        expr: sum(increase(vmalert_iteration_missed_total[15m]))
      - record: pg36:notification_failures:increase15m
        expr: sum(increase(alertmanager_notifications_failed_total[15m]))