Add rate limiting to KOI Coordinator

[DarrenZal]

Background

On Jan 9-10, 2026, the KOI pipeline experienced 57,116 false "content changed" events in 24 hours due to a bug in the GitHub sensor (fixed in regen-network/koi-sensors@b2b547e). This overwhelmed downstream services (BGE Server, OpenAI API) causing 429 rate limit errors.

Problem

The coordinator currently broadcasts all events immediately without any rate limiting. When a sensor malfunctions or a bulk re-index occurs, this can flood downstream services.

Proposed Solution

Add configurable rate limiting at the coordinator level:

# Example config
RATE_LIMIT_EVENTS_PER_MINUTE = 100
RATE_LIMIT_BURST = 50

Options to consider:

1. Token bucket - Allow bursts but enforce average rate
2. Sliding window - Hard limit per time window
3. Per-source limits - Different limits per sensor type

Implementation Considerations

• Should excess events be queued or dropped?
• Should rate limits be per-source or global?
• Need backpressure mechanism to slow down sensors?
• Config via environment variables or config file?

Related

• Event flood detection added in c3c0366
• BGE Server retry logic added in 18b197c
• Root cause (GitHub sensor RID bug) fixed in koi-sensors@b2b547e

Labels

enhancement, resilience