test(integration): Gorilla-S3 e2e — agent → MinIO → backend exact PromQL (Phase 6)

[zzylol]

Summary

End-to-end integration test for the Phase 6 (FINAL) Gorilla-S3 cold-engine pipeline:

fake-driver  ─OTLP─▶  sketchcol+gorillas3processor
                          │  encode (Gorilla XOR-delta)
                          ▼
                       MinIO   ─list/get─▶  GorillaS3ColdStore
                                                  │  decode
                                                  ▼
                                         GorillaQueryEngine
                                          exact PromQL
                                                  │
                                                  ▼
                                          backend HTTP /api/v1/query
                                          { accuracy: ε=0,
                                            data_source: gorilla_archive }

Lands integration/gorilla_s3_e2e/ mirroring the existing integration/parity/ + integration/cross_host_parity/ Go-test pattern. Validates the byte-format compatibility contract that ties together the Phase-1 Rust crate (asap-gorilla), the Phase-2 Go processor (gorillas3processor PR #282), the Phase-3 backend cold-store (PR #84), and the Phase-4/5 engine + router (PRs #85, #86).

Coordination strategy

Both of the brief's options:

• Option A (separate ports + project) — overlay uses compose project asap-gorilla-e2e on 29xxx ports so it would NOT collide with the host sweep on 19xxx, AND
• Option B (skip-if-busy) — at test setup, probe /proc/<sweep_pid> and t.Skip if alive (overridable via GORILLA_E2E_FORCE=1).

The double-defense matches the brief's "do not bring up docker unless the sweep is actually idle" gate; the live test only fires with GORILLA_E2E_LIVE=1 AND no sweep alive.

Subtests

TestGorillaS3End2End runs the live stack once and dispatches five subtests:

Subtest	Asserts
HappyPath	Chunks land in MinIO; in-test Go decoder reads the input value-set back
MultiChunkRange	≥1 chunk, with multi-chunk merge correctness if the spread crossed a 60s window
AccuracyExact	Backend response carries kind=Exact, ε=0, δ=0 (SKIPS until EngineRouter is wired into HTTP — see below)
DataSourceMarker	Backend response carries data_source: gorilla_archive (same SKIP gate)
CrossLanguageByteCompat	Go-encoded chunk decoded by the Rust asap-gorilla crate via cargo test

Plus six docker-free subtests that always run and pin the byte format / fixture invariants in CI without needing Docker (decoder round-trips, header rejection paths, fixture sum invariant).

Phase-6 follow-up surfaced by writing the test

asap-query-engine/src/drivers/query/servers/http.rs still wires Arc<SimpleEngine> directly — the Phase-5 EngineRouter is built but not yet consumed by the HTTP path. Until that lands, AccuracyExact + DataSourceMarker SKIP with a clear annotation rather than fail. Documented in PROGRESS.md and the README; tracking item for a separate ASAPQuery-backend PR.

Test plan

• go build ./... clean
• go vet ./... clean
• go test ./... passes 6 docker-free subtests + skips TestGorillaS3End2End with the documented "set GORILLA_E2E_LIVE=1" message
• In-test decoder round-trips: single-point, flat sequence, 100-sample 1..N ramp (sums to 5050, mirrors live fixture)
• GORILLA_E2E_LIVE=1 go test -v -run TestGorillaS3End2End ./... — deferred until sweep idle (sweep PID 2862786 still alive at submission time)
• File domain respected: only integration/gorilla_s3_e2e/ (new) + one PROGRESS.md line
• No submodule pointer drift; no Phase 1-5 Gorilla-S3 source touched

🤖 Generated with Claude Code