Expose metric monitor aggregates in the OpenAPI schema #129450

The public API schema does not say which aggregate values a metric monitor (detector) data source accepts. Clients such as @sentry/api and the Sentry CLI copy the list from the backend source instead, and their copies drift from it.

• BaseDetectorTypeValidator.data_sources (src/sentry/workflow_engine/endpoints/validators/base/detector.py) is a plain ListField. The generated schema therefore types it as Array<unknown>. This is still the case in @sentry/api 0.432.0.
• The only aggregates the schema mentions are prose examples in DATA_SOURCES_HELP_TEXT (workflow_engine/endpoints/validators/api_docs_help_text.py), and some are out of date. For example, it lists percentile(measurements.lcp,x), but EAP spans don't define percentile (src/sentry/search/eap/spans/aggregates.py).
• The accepted aggregates are scattered across the code:
  • SnubaQueryValidator._validate_aggregate (src/sentry/snuba/snuba_query_validator.py).
  • EAP spans/logs: EAP_FUNCTIONS, plus the definitions of each dataset (search/eap/spans, search/eap/ourlogs, which has no formulas).
  • Errors: SUPPORTED_COLUMNS / TRANSLATABLE_COLUMNS (src/sentry/incidents/logic.py).
  • Crash rate: CRASH_RATE_ALERT_AGGREGATE_RE (src/sentry/snuba/entity_subscription.py).
• The public schema is generated here (.github/workflows/openapi.yml runs make build-api-docs and copies tests/apidocs/openapi-derefed.json into getsentry/sentry-api-schema), so the change has to be made in this repo.

Impact: a client that validates or completes aggregates has to copy these rules by hand. getsentry/toolkit#1785 does this for CLI shell completion.

via @BYK.

--

View Session in Sentry