rule_files:
- alerts.yml
evaluation_interval: 1m
tests:
- name: operation failure backlog and latency fire and recover
  interval: 1m
  input_series:
  - series: trustweave_intent_operations_total{operation="reserve",outcome="storage_failure"}
    values: 0+1x15 15+0x20
  - series: trustweave_intent_reconciliation_pending
    values: 50+0x15 0+0x20
  - series: trustweave_intent_operation_seconds_bucket{operation="reserve",le="0.01"}
    values: 0+0x15 0+0x20
  - series: trustweave_intent_operation_seconds_bucket{operation="reserve",le="0.1"}
    values: 0+10x15 150+0x20
  - series: trustweave_intent_operation_seconds_bucket{operation="reserve",le="1.0"}
    values: 0+100x15 1500+0x20
  - series: trustweave_intent_operation_seconds_bucket{operation="reserve",le="5.0"}
    values: 0+200x15 3000+0x20
  - series: trustweave_intent_operation_seconds_bucket{operation="reserve",le="10.0"}
    values: 0+200x15 3000+0x20
  - series: trustweave_intent_operation_seconds_bucket{operation="reserve",le="+Inf"}
    values: 0+200x15 3000+0x20
  alert_rule_test:
  - eval_time: 0m
    alertname: IntentStorageFailure
    exp_alerts: []
  - eval_time: 10m
    alertname: IntentStorageFailure
    exp_alerts:
    - exp_labels:
        severity: critical
      exp_annotations:
        runbook: docs/operations/intent/README.md#storage-failure
  - eval_time: 35m
    alertname: IntentStorageFailure
    exp_alerts: []
  - eval_time: 0m
    alertname: IntentReconciliationBacklog
    exp_alerts: []
  - eval_time: 10m
    alertname: IntentReconciliationBacklog
    exp_alerts:
    - exp_labels:
        severity: warning
      exp_annotations:
        runbook: docs/operations/intent/README.md#reconciliation-backlog-and-age
  - eval_time: 35m
    alertname: IntentReconciliationBacklog
    exp_alerts: []
  - eval_time: 0m
    alertname: IntentAuthorizationLatency
    exp_alerts: []
  - eval_time: 10m
    alertname: IntentAuthorizationLatency
    exp_alerts:
    - exp_labels:
        severity: warning
      exp_annotations:
        runbook: docs/operations/intent/README.md#latency-and-in-flight-work
  - eval_time: 35m
    alertname: IntentAuthorizationLatency
    exp_alerts: []
- name: poll failure
  interval: 1m
  input_series:
  - series: trustweave_intent_health_poll_success
    values: 0+0x15 1+0x20
  - series: trustweave_intent_health_last_success_timestamp_seconds
    values: 0+60x36
  alert_rule_test:
  - eval_time: 0m
    alertname: IntentHealthUnavailable
    exp_alerts: []
  - eval_time: 10m
    alertname: IntentHealthUnavailable
    exp_alerts:
    - exp_labels:
        severity: critical
      exp_annotations: &id001
        runbook: docs/operations/intent/README.md#health-polling-failure-or-staleness
  - eval_time: 35m
    alertname: IntentHealthUnavailable
    exp_alerts: []
- name: poll scheduler stops despite preceding success
  interval: 1m
  input_series:
  - series: trustweave_intent_health_poll_success
    values: 1+0x36
  - series: trustweave_intent_health_last_success_timestamp_seconds
    values: 0+0x15 960+60x20
  alert_rule_test:
  - eval_time: 0m
    alertname: IntentHealthUnavailable
    exp_alerts: []
  - eval_time: 10m
    alertname: IntentHealthUnavailable
    exp_alerts:
    - exp_labels:
        severity: critical
      exp_annotations: *id001
  - eval_time: 35m
    alertname: IntentHealthUnavailable
    exp_alerts: []
- name: legacy unknown age
  interval: 1m
  input_series:
  - series: trustweave_intent_oldest_pending_known
    values: 0+0x15 1+0x20
  alert_rule_test:
  - eval_time: 0m
    alertname: IntentPendingAgeUnknown
    exp_alerts: []
  - eval_time: 10m
    alertname: IntentPendingAgeUnknown
    exp_alerts:
    - exp_labels:
        severity: warning
      exp_annotations:
        runbook: docs/operations/intent/README.md#reconciliation-backlog-and-age
  - eval_time: 35m
    alertname: IntentPendingAgeUnknown
    exp_alerts: []
- name: known pending age exceeds threshold
  interval: 1m
  input_series:
  - series: trustweave_intent_oldest_pending_age_seconds
    values: 1200+0x15 0+0x20
  alert_rule_test:
  - eval_time: 0m
    alertname: IntentPendingTooOld
    exp_alerts: []
  - eval_time: 10m
    alertname: IntentPendingTooOld
    exp_alerts:
    - exp_labels:
        severity: warning
      exp_annotations:
        runbook: docs/operations/intent/README.md#reconciliation-backlog-and-age
  - eval_time: 35m
    alertname: IntentPendingTooOld
    exp_alerts: []
- name: target scrape fails
  interval: 1m
  input_series:
  - series: up{job="trustweave-intent",instance="host-a"}
    values: 0+0x15 1+0x20
  alert_rule_test:
  - eval_time: 0m
    alertname: IntentHostUnavailable
    exp_alerts: []
  - eval_time: 10m
    alertname: IntentHostUnavailable
    exp_alerts:
    - exp_labels:
        severity: critical
        job: trustweave-intent
        instance: host-a
      exp_annotations:
        runbook: docs/operations/intent/README.md#missing-host-or-failed-scrape
  - eval_time: 35m
    alertname: IntentHostUnavailable
    exp_alerts: []
- name: target disappears from discovery
  interval: 1m
  input_series:
  - series: up{job="trustweave-intent",instance="host-a"}
    values: 1 1 stale _ _ _ _ _ _ _ _ _ _ _ _ _ 1+0x20
  alert_rule_test:
  - eval_time: 0m
    alertname: IntentHostMissing
    exp_alerts: []
  - eval_time: 10m
    alertname: IntentHostMissing
    exp_alerts:
    - exp_labels:
        severity: critical
        job: trustweave-intent
      exp_annotations:
        runbook: docs/operations/intent/README.md#missing-host-or-failed-scrape
  - eval_time: 35m
    alertname: IntentHostMissing
    exp_alerts: []
- name: scrape works but ledger instrumentation disappears
  interval: 1m
  input_series:
  - series: up{job="trustweave-intent",instance="host-a"}
    values: 1+0x36
  - series: trustweave_intent_health_poll_success{job="trustweave-intent",instance="host-a"}
    values: 1 1 stale _ _ _ _ _ _ _ _ _ _ _ _ _ 1+0x20
  - series: trustweave_intent_health_last_success_timestamp_seconds{job="trustweave-intent",instance="host-a"}
    values: 0 60 stale _ _ _ _ _ _ _ _ _ _ _ _ _ 960+60x20
  alert_rule_test:
  - eval_time: 0m
    alertname: IntentTelemetryMissing
    exp_alerts: []
  - eval_time: 10m
    alertname: IntentTelemetryMissing
    exp_alerts:
    - exp_labels:
        severity: critical
        job: trustweave-intent
        instance: host-a
      exp_annotations:
        runbook: docs/operations/intent/README.md#missing-host-or-failed-scrape
  - eval_time: 35m
    alertname: IntentTelemetryMissing
    exp_alerts: []
- name: expected outcomes do not trigger storage failure
  interval: 1m
  input_series:
  - series: trustweave_intent_operations_total{operation="reserve",outcome="success"}
    values: 0+100x36
  - series: trustweave_intent_operations_total{operation="reserve",outcome="rejected"}
    values: 0+100x36
  - series: trustweave_intent_operations_total{operation="reserve",outcome="invalid"}
    values: 0+100x36
  - series: trustweave_intent_operations_total{operation="reserve",outcome="cancelled"}
    values: 0+100x36
  alert_rule_test:
  - eval_time: 10m
    alertname: IntentStorageFailure
    exp_alerts: []
- name: partial telemetry loses timestamp and another host cannot mask it
  interval: 1m
  input_series:
  - series: up{job="trustweave-intent",instance="host-a"}
    values: 1+0x36
  - series: up{job="trustweave-intent",instance="host-b"}
    values: 1+0x36
  - series: trustweave_intent_health_poll_success{job="trustweave-intent",instance="host-a"}
    values: 1+0x36
  - series: trustweave_intent_health_poll_success{job="trustweave-intent",instance="host-b"}
    values: 1+0x36
  - series: trustweave_intent_health_last_success_timestamp_seconds{job="trustweave-intent",instance="host-a"}
    values: 0 60 stale _ _ _ _ _ _ _ _ _ _ _ _ _ 960+60x20
  - series: trustweave_intent_health_last_success_timestamp_seconds{job="trustweave-intent",instance="host-b"}
    values: 0+60x36
  alert_rule_test:
  - eval_time: 0m
    alertname: IntentTelemetryMissing
    exp_alerts: []
  - eval_time: 1m
    alertname: IntentTelemetryMissing
    exp_alerts: []
  - eval_time: 3m
    alertname: IntentTelemetryMissing
    exp_alerts: []
  - eval_time: 5m
    alertname: IntentTelemetryMissing
    exp_alerts:
    - exp_labels:
        severity: critical
        job: trustweave-intent
        instance: host-a
      exp_annotations:
        runbook: docs/operations/intent/README.md#missing-host-or-failed-scrape
  - eval_time: 16m
    alertname: IntentTelemetryMissing
    exp_alerts: []
  - eval_time: 35m
    alertname: IntentTelemetryMissing
    exp_alerts: []