clusterloader2/testing/load/modules/dns-performance-metrics.yaml

# Valid actions: "start", "gather"
{{$action := .action}}

{{$ENABLE_DNSTESTS := DefaultParam .CL2_ENABLE_DNSTESTS false}}
# Guard the new DNS tests. Remove it once it's confirmed that it works on a subset of tests.
{{$USE_ADVANCED_DNSTEST := DefaultParam .CL2_USE_ADVANCED_DNSTEST false}}
# DNS test threshold parameters.
{{$DNS_ERROR_PERC_THRESHOLD := DefaultParam .CL2_DNS_ERROR_PERC_THRESHOLD 0.1}}
{{$DNS_LOOKUP_LATENCY_50_THRESHOLD := DefaultParam .CL2_DNS_LOOKUP_LATENCY_50_THRESHOLD 0.02}}
{{$DNS_LOOKUP_LATENCY_99_THRESHOLD := DefaultParam .CL2_DNS_LOOKUP_LATENCY_99_THRESHOLD 0.1}}

{{if and $ENABLE_DNSTESTS $USE_ADVANCED_DNSTEST}}
steps:
- name: "{{$action}}ing measurements"
  measurements:
  - Identifier: DNSPerformanceMetrics
    Method: GenericPrometheusQuery
    Params:
      action: {{$action}}
      metricName: DNS Performance
      metricVersion: v1
      unit: s
      enableViolations: true
      queries:
      - name: DNS Lookup Count
        query: sum(increase(dns_lookups_total[%v]))
      - name: DNS Timeout Count
        query: sum(increase(dns_timeouts_total[%v]))
      - name: DNS Error Count
        query: sum(increase(dns_errors_total[%v]))
      - name: DNS Error Percentage
        query: sum(increase(dns_errors_total[%v])) / sum(increase(dns_lookups_total[%v])) * 100
        threshold: {{$DNS_ERROR_PERC_THRESHOLD}}
      - name: DNS Lookup Latency - Perc50
        query: histogram_quantile(0.5, sum(rate(dns_lookup_latency_bucket[%v])) by (le))
        threshold: {{$DNS_LOOKUP_LATENCY_50_THRESHOLD}}
      - name: DNS Lookup Latency - Perc99
        query: histogram_quantile(0.99, sum(rate(dns_lookup_latency_bucket[%v])) by (le))
        threshold: {{$DNS_LOOKUP_LATENCY_99_THRESHOLD}}
{{end}}