fix(emit): cumulative streaming-config across (metric, role) pairs (B2 follow-up)

[zzylol]

Summary

PR #283 made WorkloadStore and PlanStore (metric, role)-keyed so a single metric can carry multiple aggregation roles (e.g. http_requests_total post-B2 has both DDSketch-Quantile from quantile_over_time(...) AND ExactAgg-Sum from sum by (zone) (http_requests_total)). But the streaming-config emit path in control_plane/src/main.rs was still metric-keyed and emitted the current iteration's single-role BackendStageConfig alone.

The data plane's POST /api/v1/streaming-config handler is an atomic full handle.swap(new_config) (see data_plane/src/drivers/query/servers/http.rs:4758), so the second per-role POST destroys the first role's aggregations on the backend → sum by (zone) (http_requests_total) returns ExactAgg(Sum) capability not satisfied.

Changes (all in control_plane/src/main.rs)

• Rekey backend_routing_cache from HashMap<String, BackendStageConfig> to HashMap<(String, AggRole), BackendStageConfig>. Imports AggRole from control_plane::workload.
• Insert with role: uses the role derived just above in handle_plan (already in scope via derive_agg_role on a synthesized WorkloadEntry from the request).
• Cumulative streaming-config emit: builds one BackendStageConfig by concatenating aggregations + readouts across every cache entry, then passes that single cumulative config to emit_backend_streaming_config_json. The data plane's swap installs every role's aggregations atomically.
• Per-metric merge for storage-routing: groups cache entries by metric name and merges each metric's BackendStageConfigs (concat aggregations + readouts) before passing to emit_backend_storage_routing. The routing classifier reads cfg.aggregations to derive shape routing, so a metric carrying both DDSketch + Sum now routes both shape families correctly.
• Doc update on backend_routing_cache reflecting the new (metric, role) keying and cumulative-across-all-roles emit semantic.
• Deterministic ordering: cache iteration sorts by metric then role so emitted JSON is reproducible across runs (avoids flaky regression assertions on captured bodies).

Constraints honored: no data_plane/ changes, no signature changes to emit_backend_streaming_config_json or emit_backend_storage_routing (merge happens at the call site), fire-and-forget POST contract preserved.

Regression test

api_tests::streaming_config_cumulative_push_covers_all_planned_metrics — replays the demo's 5-metric plan-POST sequence against a mock backend, captures every streaming-config body, asserts the LAST body (the one the data plane's atomic swap installs) carries aggregations for ALL 5 metrics. Pre-fix this fails with "missing aggregations for metric X" because only the most recent plan-POST survives.

Sibling to the existing storage_routing_cumulative_push_covers_all_5_sketched_metrics test which covers the storage-routing analogue.

Test results

• cargo build -p control_plane — clean.
• cargo test -p control_plane --lib — 740 passed, 0 failed.
• cargo test -p control_plane --bin control_plane — 28 passed (including the new regression test and the existing storage_routing cumulative test).

Smoke test — partial result, separate root cause

The MVP smoke test does NOT exercise this fix because the smoke harness never POSTs to /api/v1/plan — it relies on the controller's startup pre-pop loop + agent-connect path, neither of which calls the typed Backend-stage emit. As a result:

• The mock-backend GET on /api/v1/streaming-config returns the STATIC YAML the backend loaded from backend-streaming.yaml at boot (two DDSketch entries for http_requests_total and http_requests_total_latency_ms), not anything the controller posted.
• quantile_over_time(0.99, http_requests_total_latency_ms[5m]) PASSES (4 zones with values).
• sum by (zone) (http_requests_total) returns "No result for query" because the backend's http_requests_total aggregation is statically DDSketch, NOT ExactAgg(Sum) — the controller never pushed an ExactAgg(Sum) entry over it.

This is a separate gap (controller pre-pop loop + agent-connect path don't post cumulative streaming-config). My PR fixes the handle_plan cumulative emit path the task describes — it's correct + tested in isolation, but the smoke harness exercises a different code path that needs its own follow-up. Not regressing anything that previously worked.

Test plan

• cargo build clean
• cargo test -p control_plane --lib + --bin (768 tests passed)
• New regression test covers cumulative streaming-config across plan POSTs
• Smoke test sum by (zone) (http_requests_total) — blocked by separate gap (controller's pre-pop + agent-connect paths don't trigger cumulative streaming-config push to backend; this PR only fixes the handle_plan HTTP POST path)

🤖 Generated with Claude Code