Output metric ADR

Signed-off-by: David Jetelina <[email protected]>

adr/0001_rule_outputs.md

@@ -0,0 +1,75 @@
+# Output metric naming
+
+* Status: proposed
+* Date: 2023-10-27
+
+## Context and Problem Statement
+
+The solution provided by the operator doesn't end with a `PrometheusRule` CRD, but takes into account as much of the SLO journey as possible. 
+Part of that journey is displaying Indicators, Objectives, Error Budgets etc. on different dashboards. The design should account for
+existence of reusable Grafana dashboards at the very least.
+
+## Considered Options
+
+* Prometheus Recording Rule naming convention
+  * With part of slo in name: `slo:logging_query_frontend_error_rate:error_budget:28d`
+  * With osko in name: `osko:error_budget{slo_name="logging_query_frontend_error_rate", window="28d", ...}`
+* Pretend to be a metric
+  * `osko_error_budget{slo_name="logging_query_frontend_error_rate", window="28d", ...}`
+
+## Decision Outcome
+
+Chosen option: "Pretend to be a metric". There is a reasonable chance that `PrometheusRule` CRD might not be the only possible output
+of OSKO and thus it's worth it to avoid the result being too implementation specific.
+
+The metric names that should be exposed/recorded:
+
+* `osko_sli_ratio_good`
+* `osko_sli_ratio_bad`
+* `osko_sli_ratio_total`
+* `osko_sli_measurement`
+* `osko_slo_target`
+* `osko_error_budget_available`
+
+Ideally also the following, although these might be implementation specific (aka we'll see how hard it is):
+
+* `osko_error_budget_burn_rate`
+* `osko_error_budget_burn_rate_threshold`
+
+The following labels should be present next to the metric (where applicable):
+
+*Note that the following might change as we explore implementing composite SLOs and other more complex scenarios.*
+
+* `sli_name`
+* `slo_name`
+* `service`
+* `window`
+
+### Positive Consequences
+
+* Future implementations will be able to plug in to the rest of the solution
+* Exhaustive list of metrics will make creation of the dashboards easy
+
+### Negative Consequences
+
+* Tracing resulting metric back to the original will be only possible by looking at the SLI specification manifest.
+
+## Pros and Cons of the Options
+
+### Prometheus Recording Rule naming convention - With part of slo in nam
+
+`slo:loki_request_duration_seconds_count:ratio_good:28d`
+
+* Good, because it can show the original metric
+* Bad, because dashboards would have to be generated for each metric
+
+### Prometheus Recording Rule naming convention -  With osko in name
+
+`osko:error_budget{slo_name="logging_query_frontend_error_rate", window="28d", ...}`
+
+* Good, because it allows for reusable dashboards
+* Bad, because it's tightly tied to Prometheus recording rule implementation
+
+## Links 
+
+* [Prometheus recordingrule naming recommendations](https://prometheus.io/docs/practices/rules/#naming)