Return Errors for QueryAPI when Upstream is Down

[hanzhang911]

Our Proposed Architecture

We are building a new architecture using Thanos for monitoring. Due to the very large amount of active series, we plan to host different categories of data in different shards.
On each shard, we deploy a Thanos querier to talk to the upstream Mimir. We also use a separate Thanos querier to federate the results returned from these shards. So it looks like this

Returning partial result or failure when one upstream querier is down

We are exploring how the above Thanos-federation-querier behaves (return partial results or failures/erros) when one of the upstream queriers (querier on shard-1 or on shard-2) is down. We have not decided what to return, but we wanted to see what are the options that are natively supported by Thanos.
We found that the arguments named “--query.partial-response” and “--no-query.partial-response” can control the behavior. However, further tests show that these two arguments are only effective for StoreAPI. And they do not apply to QueryAPI.
In our case, the federation-querier is talking to the two upstream queriers via QueryAPI, so even if we specify “--no-query.partial-response”, the federation-querier still returns partial results, instead of returning failures.

We have also tried pass the HTTP parameter partial_response=false as follows when querying the federation querier, but it still returns partial results instead of errors.

curl -G 'http://localhost:3008/api/v1/query_range' \
  --data-urlencode 'query=sum(cpu)' \
  --data-urlencode 'start=2025-04-30T18:55:00Z' \
  --data-urlencode 'end=2025-04-30T18:57:00Z' \
  --data-urlencode 'partial_response=false' \
  --data-urlencode 'step=10s' -H 'X-Scope-OrgID: default'

So my question is that

1. If QueryAPI is being used, is there a way to ask the federation-querier return failures/errors when at least one of the upstream queriers are down?
2. If the answer to question (1) is no, is there a plan to support this feature?

Thanks
Han