feat: standard OpAMP protobuf + e2e integration status

[zzylol]

Summary

Rewrites the controller's OpAMP WebSocket server to speak the standard OpAMP
protobuf protocol. Adds OpAMP extension and S3 exporter to the sketchcol builder.

E2E Integration Test Results (real test, not simulated)

Step	What	Status	Evidence
1. Controller starts	Rust binary, HTTP API on :8080, OpAMP on :4320	✅	curl /api/v1/agents returns {}
2. Collector connects	countminsketchcol with opampextension → controller WS	✅	Controller returns {"test-agent-1":"agent"}
3. Plan submitted	POST /api/v1/plan → planner → YAML → OpAMP push	✅	agents_notified: 1, sketch_type: KLL
4. Controller sends protobuf	ServerToAgent { remote_config } binary frame	✅	Unit tests verify encode/decode round-trip
5. Collector receives config	opampextension decodes ServerToAgent	❌ not supported	Extension logs show no config received
6. Collector applies config	Pipeline restarts with new sketch processor	❌ not supported	opampextension lacks AcceptsRemoteConfig

Root cause

The opampextension in opentelemetry-collector-contrib does not implement
AcceptsRemoteConfig. It only supports:

• ReportsEffectiveConfig — reports current config to server
• ReportsHealth — reports health status
• ReportsAvailableComponents — reports available components

It does NOT support:

• AcceptsRemoteConfig — receive and apply config from server
• Config hot-reload based on server push

This is a known limitation of the upstream extension. The OpAMP spec defines
AcceptsRemoteConfig as a capability, but the Go implementation doesn't have it.

Potential solutions

Solution	Effort	How it works	Pros	Cons
A. OpAMP Supervisor	Low	Use opamp-go supervisor — a separate process that manages the collector, receives OpAMP remote config, writes it to a file, and restarts the collector	Standard OpAMP approach; no fork needed; collector binary unchanged	Extra process to manage; restart latency (~seconds)
B. Fork opampextension	Medium	Add AcceptsRemoteConfig to the Go extension — decode ServerToAgent.remote_config, write YAML to confmap, trigger collector reload	In-process; no restart; fastest config application	Requires maintaining a fork of opentelemetry-collector-contrib
C. HTTP polling	Low	Collector uses httpprovider with a watcher that polls GET /api/v1/config/:metric periodically; controller updates the config, collector re-fetches	No OpAMP needed; works with any collector binary	Polling delay (seconds); not push-based; collector must know which metric to poll
D. File-based + fsnotify	Low	Controller writes YAML to a shared volume; collector uses fileprovider with --config file:config.yaml; fsnotify triggers reload on file change	Simple; works today with OTel's built-in file watching	Requires shared filesystem; not suitable for distributed deployments
E. Contribute upstream	High	Submit PR to opentelemetry-collector-contrib adding AcceptsRemoteConfig to opampextension	Benefits entire community; no fork maintenance	Long review cycle; may not align with upstream priorities

Recommended path: Solution A (OpAMP Supervisor) for immediate use, then Solution E
(upstream contribution) for long-term. The supervisor approach is the officially
recommended way to use OpAMP for remote config in the OTel ecosystem.

What this PR still delivers (works today)

1. Standard OpAMP protobuf on the wire — future-proof for when AcceptsRemoteConfig is available
2. OpAMP connection establishment — controller sees agents, can identify roles
3. httpprovider bootstrap — collectors start with controller-generated config
4. S3 dual-path config — sketch pipeline + raw backup (paper §3 Fig.1)
5. Per-metric HTTP config — GET /api/v1/config/:metric serves YAML on demand

Files changed (8 files, +1382 −18)

File	What
proto/opamp.proto (+1079)	Standard OpAMP protobuf spec
proto/anyvalue.proto (+67)	OpAMP dependency
build.rs (+7)	prost-build compiles proto
Cargo.toml (+5)	+prost, +bytes, +prost-build
src/opamp/mod.rs (+106 −18)	Protobuf ServerToAgent send, AgentToServer receive
sketchcol/builder-config.yaml (+8 −1)	+opampextension, +awss3exporter
sketchcol/config-dual-path.yaml (+76)	Sketch + S3 dual pipeline
sketchcol/config-with-opamp.yaml (+52)	OpAMP-connected collector config

Tests

• 329 pass total
• push_to_role_delivers_yaml_to_agent_role — protobuf binary encode → WebSocket → decode → verify YAML body + hash
• push_to_agent_role_does_not_reach_backend_role — role-based routing verified with protobuf

🤖 Generated with Claude Code