Summary
Add backpressure mechanisms to the transaction submission path to handle DA layer degradation gracefully.
Parent: #4
Context
celestia-node has zero rate limiting or backpressure on submission — just a mutex serializing broadcasts. Under load or during network issues, this leads to unbounded retries and resource exhaustion.
Requirements
Concurrency limiter
Exponential backoff
- On repeated submission failures, increase delay between attempts
- Base: 1s, max: 30s, jitter: +/- 25%
- Reset backoff on successful submission
Circuit breaker
- Track failure rate over a sliding window (e.g., last 20 submissions)
- States: closed (normal) → open (failing, reject immediately) → half-open (probe)
- Open threshold: >50% failure rate over window
- Half-open: allow 1 submission, if success → closed, if fail → open
- When open: return immediate error with "circuit open" context, don't waste gas
Metrics (#7)
apex_submission_inflight (gauge)
apex_submission_circuit_state (gauge — 0=closed, 1=half-open, 2=open)
apex_submission_backoff_seconds (gauge)
[submission.rate_limit]
max_inflight = 4
backoff_base = "1s"
backoff_max = "30s"
circuit_window = 20
circuit_open_threshold = 0.5
References
- Sony's gobreaker or similar for circuit breaker pattern