service-level-operator
service-level-operator copied to clipboard
Add burn rate threshold levels for the SLO
The alerts based on burn rate thresholds can be made easier if the operator exposes metrics based on the CRD thresholds.
My idea at this moment is having something like this on the CRD:
apiVersion: measure.slok.xyz/v1alpha1
kind: ServiceLevel
metadata:
name: awesome-service
spec:
serviceLevelObjectives:
# A typical 5xx request SLO.
- name: "9999_http_request_lt_500"
description: 99.99% of requests must be served with <500 status code.
disable: false
availabilityObjectivePercent: 99.99
burnRates:
- errorBudgetDays: 30
thresholds:
- timeRangeHours: 1
errorBudgetPercent: 2
- timeRangeHours: 6
errorBudgetPercent: 5
- timeRangeHours: 72
errorBudgetPercent: 10
serviceLevelIndicator:
prometheus:
address: http://127.0.0.1:9091
totalQuery: |
sum(
increase(skipper_serve_host_duration_seconds_count{host="www_spotahome_com"}[2m]))
errorQuery: |
sum(
increase(skipper_serve_host_duration_seconds_count{host="www_spotahome_com", code=~"5.."}[2m]))
output:
prometheus: {}
We could have multiple burnRates and in each burn rate multiple thresholds.
I have a branch that creates the threshold metrics and sets the threshold information on labels:
# HELP service_level_slo_burn_rate_threshold Is the threshold for a burn rate period.
# TYPE service_level_slo_burn_rate_threshold gauge
service_level_slo_burn_rate_threshold{burn_rate_range="168h",error_budget_spent="10%",fake="true",namespace="ns0",service_level="fake-service0",slo="fake_slo0",team="fake-team0",total_error_budget_range="70d"} 1
service_level_slo_burn_rate_threshold{burn_rate_range="1h",error_budget_spent="2%",fake="true",namespace="ns0",service_level="fake-service0",slo="fake_slo0",team="fake-team0",total_error_budget_range="30d"} 14.4
service_level_slo_burn_rate_threshold{burn_rate_range="1h",error_budget_spent="2%",fake="true",namespace="ns0",service_level="fake-service0",slo="fake_slo1",team="fake-team1",total_error_budget_range="30d"} 14.4
service_level_slo_burn_rate_threshold{burn_rate_range="1h",error_budget_spent="2%",fake="true",namespace="ns0",service_level="fake-service0",slo="fake_slo2",team="fake-team2",total_error_budget_range="30d"} 14.4
service_level_slo_burn_rate_threshold{burn_rate_range="24h",error_budget_spent="7%",fake="true",namespace="ns0",service_level="fake-service0",slo="fake_slo0",team="fake-team0",total_error_budget_range="70d"} 4.9
service_level_slo_burn_rate_threshold{burn_rate_range="6h",error_budget_spent="3%",fake="true",namespace="ns0",service_level="fake-service0",slo="fake_slo0",team="fake-team0",total_error_budget_range="70d"} 8.4
service_level_slo_burn_rate_threshold{burn_rate_range="6h",error_budget_spent="5%",fake="true",namespace="ns0",service_level="fake-service0",slo="fake_slo0",team="fake-team0",total_error_budget_range="30d"} 6
service_level_slo_burn_rate_threshold{burn_rate_range="6h",error_budget_spent="5%",fake="true",namespace="ns0",service_level="fake-service0",slo="fake_slo1",team="fake-team1",total_error_budget_range="30d"} 6
service_level_slo_burn_rate_threshold{burn_rate_range="6h",error_budget_spent="5%",fake="true",namespace="ns0",service_level="fake-service0",slo="fake_slo2",team="fake-team2",total_error_budget_range="30d"} 6
service_level_slo_burn_rate_threshold{burn_rate_range="72h",error_budget_spent="10%",fake="true",namespace="ns0",service_level="fake-service0",slo="fake_slo0",team="fake-team0",total_error_budget_range="30d"} 1
service_level_slo_burn_rate_threshold{burn_rate_range="72h",error_budget_spent="10%",fake="true",namespace="ns0",service_level="fake-service0",slo="fake_slo1",team="fake-team1",total_error_budget_range="30d"} 1
service_level_slo_burn_rate_threshold{burn_rate_range="72h",error_budget_spent="10%",fake="true",namespace="ns0",service_level="fake-service0",slo="fake_slo2",team="fake-team2",total_error_budget_range="30d"} 1
# HELP service_level_slo_objective_ratio Is the objective of the SLO in ratio unit.
Any thoughs? @ese