fix(deploy): remove asap-gateway (no longer needed post-wave; saves ~10 Mbps double-hop)

[zzylol]

Rationale

The asap-gateway OTel-collector hop between agents and the asap-backend was originally the cross-agent fan-in tier for sum by (zone) (...) aggregations. After ASAPQuery-backend PRs #283 (B2 stage-emit) and #287/#290 (cumulative emit), the wave-tier backend does that fan-in natively via its evaluate_exact_agg reducer (pulls sids from every agent at query time and merges across hosts).

The gateway became a pure OTLP forwarder that doubled every wire byte (agent → gateway → backend) for no algorithmic reason, costing ~10 Mbps per arm on the multinode demo. See issue #400.

Scope

• deploy/mvp-multinode/scripts/run_demo.sh — drop gateway_up / gateway_down; remove the gateway from arm_up / arm_down. stop_node node1 stays in teardown so pre-asap pipeline uses ~4.3× more bandwidth than baseline (multi-family fan-out + double-hop) #400 leftover containers get reaped.
• deploy/mvp-multinode/topology.env — drop the --add-host=gateway: alias.
• deploy/mvp-multinode/configs/asap/asap-otel-agent-b6-asap-single-sketch.yaml + singlenode equivalent — OTLP endpoint: flips from gateway:4317 to backend:4317.
• deploy/mvp-{multi,single}node/configs/.../mvp-workload.yaml — the lone assign_to_role: gateway entry flips to assign_to_role: agent. With no Gateway stage in the DAG, the controller's typed split_typed_three_stage simply doesn't emit a Gateway stage and push_to_role(Gateway, ...) is a no-op. No control_plane code change needed.
• deploy/mvp-multinode/configs/asap/asap-otel-gateway-mvp-placeholder.yaml — deleted.
• deploy/mvp-singlenode/docker-compose/base.yml — gateway service removed; host ports 14317 / 14318 reassigned to backend. fake-exporter EXPORTER_TARGET defaults to backend:4317; depends_on updated.
• deploy/mvp-singlenode/docker-compose/agents-N{1,10,100}.yml + gen-agents.sh — depends_on: gateway → depends_on: backend.

Validation (multinode, asap arm, WARMUP_S=90)

docker ps after run_demo.sh up asap:

node	containers
node0	agent-a + 5 producers
node1	(empty — no asap-gateway)
node2	backend + controller + minio + prometheus + thanos-{query,store-gateway,compact}
node3	agent-b + 5 producers

All 5 wave queries against http://10.10.1.3:9091/api/v1/query returned data_source: asap_query:

query	n_series	sample value
quantile_over_time(0.99, http_requests_total_latency_ms[5m])	10000	per-series p99
max by (zone) (quantile_over_time(0.99, http_requests_total_latency_ms[5m]))	4	z0 = 165.70 ms
sum by (zone) (http_requests_total)	4	z0 = 14888877
sum by (zone) (rate(http_requests_total[5m]))	4	z0 = 159509.04
topk(5, sum by (zone) (rate(http_requests_total[5m])))	4	z1 = 199208.09

Bandwidth before / after

NIC: enp130s0f0, 60 s sample during steady-state asap arm.

node	role	RX (Mbps)	TX (Mbps)
node0	agent-a → backend	0.16	6.80
node1	(unused; was gateway)	0.00	0.00
node2	backend (receives both agents)	13.56	0.33
node3	agent-b → backend	0.16	6.76

Pre-fix the node1 link was carrying ~13 Mbps in + ~13 Mbps out as a pure forwarder (the double-hop the gateway added). Post-fix that ~26 Mbps of node1 wire traffic is gone; the only cross-host bytes are agent→backend at 13.56 Mbps total.

Test plan

• run_demo.sh sync succeeds
• run_demo.sh up asap succeeds with no gateway bring-up
• node1 docker ps shows no asap-gateway
• node2 docker ps shows the expected backend stack with no gateway
• All 5 wave queries return non-empty results via data_source: asap_query
• run_demo.sh down cleanly tears down all 4 nodes
• Steady-state aggregate bandwidth ≤ 14 Mbps (was ~27 Mbps with the gateway double-hop)

🤖 Generated with Claude Code