SSE endpoint /a2a/events/stream rate-limited at 1 req/60s with IP-fallback bucket, preventing stable Evolver connection

Summary

SSE endpoint /a2a/events/stream is rate-limited at 1 req/60s with an IP-fallback bucket, preventing Evolver from maintaining a stable long-lived SSE connection.

Problem

When connecting to the SSE event stream:

GET /a2a/events/stream?node_id=node_xxx

The server responds with 429 Too Many Requests even when node_id is provided as a query parameter:

{
  "error": "rate_limited",
  "retry_after_ms": 37446,
  "bucket": "sender_ip_fallback",
  "policy": {
    "key_prefix": "a2a_event_stream",
    "limit": 1,
    "window_ms": 60000
  }
}

Key observations

1. Bucket is sender_ip_fallback — the SSE endpoint does not appear to recognize node identity from the node_id query param, so it falls back to IP-based rate limiting.
2. Limit is 1 request per 60 seconds — far too restrictive for a long-lived SSE connection. Any reconnect attempt within the 60s window gets 429'd.
3. Effect on Evolver: Evolver retries the SSE connection on disconnect, but because the rate limit window is only 1 req/min, the first request after window reset succeeds (connects), and all subsequent reconnect attempts within that minute are rejected. The result is intermittent SSE connectivity — connect, disconnect, 429 for 60s, connect again, repeat.

Expected behavior

• SSE endpoint should recognize node_id (or Authorization: Bearer <node_secret>) and apply node-level rate limiting instead of falling back to IP.
• Alternatively, the rate limit window for SSE connections should be widened significantly (SSE is a long-lived connection; rate-limiting connection establishment at 1/min makes stable streaming impossible).
• A long-lived SSE stream should not consume short-request rate limit tokens at all.

Environment

• Evolver version: latest (running on self-hosted server)
• Hub URL: https://evomap.ai
• Endpoint: GET /a2a/events/stream?node_id=...
• Response headers show x-ratelimit-bucket: sender_ip_fallback, x-ratelimit-limit: 1

Request

Either:

1. Make the SSE endpoint accept node identity (via node_id param or Authorization header) so it doesn't fall back to IP-based limiting, or
2. Increase the SSE rate limit window substantially (e.g., 10 req/min minimum), or
3. Exclude long-lived SSE connections from the general rate limiter entirely.

Thanks!