QueryRange: Catalogs range validation as an internal error

[gotjosh]

Time ranges for range queries are validated against a per-tenant limit.

cortex/pkg/querier/querier.go

Lines 285 to 289 in e0663b7

	startTime := model.Time(sp.Start)
	endTime := model.Time(sp.End)
	if maxQueryLength := q.limits.MaxQueryLength(userID); maxQueryLength > 0 && endTime.Sub(startTime) > maxQueryLength {
	return storage.ErrSeriesSet(fmt.Errorf(validation.ErrQueryTooLong, endTime.Sub(startTime), maxQueryLength))
	}

Within Prometheus API, the query_range endpoint does all of its validation upfront - meaning that incorrect input errors (400s) are validated before the query is ran.

https://github.com/prometheus/prometheus/blob/a282d2509971174301408c5a5f96946c478dfb0f/web/api/v1/api.go#L387-L452

As it is, Cortex will return a 500 for the validation of this limit which seems incorrect. Any thoughts?

I'm not sure upstream will accept changes to encapsulate custom validation as part of the API, the other options is to copy the validation code of the code of this particular endpoint which seems unideal.

An example of the error:

{
  "status": "error",
  "errorType": "internal",
  "error": "expanding series: the query time range exceeds the limit (query length: 4414h5m0s, limit: 768h0m0s)"
}

[pstibrany]

No need to change upstream, we can already convert this error to 422. This happens in https://github.com/cortexproject/cortex/blob/527e1c41b72081d1d43f986ac8a64622c3f05500/pkg/api/queryable.go, which understands some error types and returns appropriate promql.Err* expected by Prometheus API (which then translates it to HTTP status code)