feat (thanos-ASAPCollector): implement /api/v1/query_range forwarding in ThanosQueryEngine

[SieDeta]

Summary

This PR implements /api/v1/query_range forwarding in ThanosQueryEngine so that range queries against archive-tier metrics are served by Thanos instead of returning an error.

---

What Was Added

• Added range_endpoint() to ThanosQueryConfig

  • Builds:

    {base_url}/api/v1/query_range
    

• Added query_range() to ThanosQueryEngine

  • POSTs the query to Thanos
  • Forwards:
    • query
    • start
    • end
    • step
  • Converts timestamps from milliseconds to Unix seconds

• Added execute_range() override on ThanosQueryEngine

  • Maps:
    • 4xx → CapabilityMiss
    • 5xx / timeout → Backend

• Added execute_range() default method to the QueryEngine trait

  • Returns CapabilityMiss by default
  • Existing engine implementations remain unchanged

• Wired process_range_query_request() in http.rs

  • Falls through to ThanosQueryEngine when:

    ASAPQueryEngine::execute_range_promql_modern()

    returns:

    CapabilityMiss

• Added:

  • 6 unit tests
  • 3 mock helpers
  • 1 HTTP e2e test

• Updated:

  • verify_gorilla_compression.sh
  • README.md

---

Why

ThanosQueryEngine previously only forwarded instant queries through:

POST /api/v1/query

Range queries still went through:

ASAPQueryEngine::execute_range_promql_modern()

When the ASAP sketch tier had no warm data for the metric, the backend returned:

unsupported query

instead of falling through to Thanos.

This blocked:

• Grafana panels using query_range
• Direct /api/v1/query_range requests
• Archive-tier / Gorilla-backed metric queries

even though Thanos already had the data available via:

• thanos-store-gateway
• gorilla-buffer-store

The ingest path and instant-query path are unchanged.

---

Flow Before

GET /api/v1/query_range
    -> ASAPQueryEngine::execute_range_promql_modern
    -> CapabilityMiss (no sketch data)
    -> 404 "unsupported query"

---

Flow After

GET /api/v1/query_range
    -> ASAPQueryEngine::execute_range_promql_modern
    -> CapabilityMiss (no sketch data)
    -> ThanosQueryEngine::execute_range
    -> POST thanos-query:10903/api/v1/query_range
    -> 200 matrix response

---

Test Results

Unit Tests

cargo test --manifest-path data_plane/Cargo.toml thanos_query_engine

test query_range_calls_range_endpoint_not_instant  ... ok
test query_range_passes_params_correctly           ... ok
test query_range_parses_matrix_response            ... ok
test query_range_4xx_returns_capability_miss       ... ok
test query_range_5xx_returns_backend_error         ... ok
test query_range_timeout_returns_backend_error     ... ok
... 11 pre-existing tests                          ... ok

test result: ok. 17 passed; 0 failed

HTTP End-to-End Test

cargo test http_query_range_forwards_to_thanos

test http_query_range_forwards_to_thanos_when_asap_misses ... ok

test result: ok. 1 passed; 0 failed

[zzylol]

Superseded by #347.

The Thanos /api/v1/query_range forwarding from this PR was carried forward — the implementation (forward.rs query_range/execute_range + tests) was ported into #347, thanks for it.

The difference is structural: a routing review concluded engine selection should live in one place (EngineRouter) rather than being inlined in http.rs. So #347 adds EngineRouter::execute_range (plus the QueryEngine::execute_range trait method and ASAPQueryEngine/ThanosQueryEngine overrides) and routes both instant and range queries uniformly through the router — ASAP/sketch tier first, archive (Thanos) fallback on CapabilityMiss — keeping http.rs a thin transport layer.

Closing in favor of #347.