feat: push-based config delivery — controller → OpAMP Supervisor → collector (#132)

[zzylol]

Closes #132.

E2E Test Results — VERIFIED WORKING

Full push-based config delivery tested and confirmed working on this machine:

Step	What	Status	Evidence
1. Controller starts	HTTP :8080, OpAMP WS :4320	✅	curl /api/v1/agents → {}
2. Supervisor connects	OpAMP WebSocket	✅	"Connected to the server." + {"cms-agent-1":"agent"}
3. Plan pushed	CountMinSketch, window mode, 1m	✅	sketch_type: countmin, agents_notified: 1
4. Supervisor receives	Saves + merges config	✅	effective.yaml shows countmin processor
5. Collector applies	Starts with pushed config	✅	"Starting Count-Min Sketch processor" mode: window, window_duration: 60
6. Collector running	Accepts OTLP metrics	✅	OTLP HTTP endpoint responding, PID confirmed

Pushed config applied by collector

processors:
    countmin:
        columns: 2048
        rows: 5
        metric_name: countsketch_partition
        mode: window
        window_duration: 1m
        transmit_sketch: false
        enable_self_monitoring: true

This config was generated by the controller (generate_agent_config()), pushed via OpAMP (ServerToAgent protobuf), received by the supervisor, and applied to the running collector.

Architecture

Controller (Rust)                OpAMP Supervisor (Go)              Collector (Go)
  │                                   │                                 │
  │  POST /api/v1/plan                │                                 │
  │  → planner → CountMinSketch       │                                 │
  │  → generate_agent_config() → YAML │                                 │
  │  → ServerToAgent protobuf         │                                 │
  │  ─── WebSocket binary frame ────▶ │                                 │
  │                                   │  save remote config             │
  │                                   │  merge with local config        │
  │                                   │  write effective.yaml           │
  │                                   │  ─── start/restart collector ─▶ │
  │                                   │                                 │  countmin processor
  │                                   │                                 │  window mode, 1m
  │                                   │                                 │  OTLP :4317/:4318

How to reproduce

# 1. Start controller
cargo run --release -p controller

# 2. Build supervisor (one-time)
cd opentelemetry-collector-contrib/cmd/opampsupervisor && go build -o opampsupervisor .

# 3. Start supervisor
./opampsupervisor --config supervisor-config.yaml

# 4. Push a plan — collector gets new config automatically
curl -X POST http://localhost:8080/api/v1/plan \
  -H "Content-Type: application/json" \
  -d '{"metric_name":"cpu","aggregations":["Frequency"],"time_window":"1m","accuracy_sla":0.01,"sketch_type":"countminsketch"}'

# 5. Verify collector is running with pushed config
curl -X POST http://localhost:4318/v1/metrics -H "Content-Type: application/json" -d '{}'
# → {"partialSuccess":{}} means collector is accepting metrics

Files

File	What
sketchcol/supervisor-config.yaml	OpAMP Supervisor config connecting to controller

🤖 Generated with Claude Code