Rates Engine v0.5.0-rc.69

[v0.5.0-rc.69] — 2026-05-22

Added

• /v1/ledger/tip + /v1/ledger/stream — the live-ingest
  frontier (#58). /v1/ledger/tip is a lightweight endpoint
  returning just the highest ledger the indexer has committed and
  its lag — a status page or monitor can poll "what ledger are we
  on" without pulling the whole /v1/diagnostics/ingestion
  snapshot. latest_ledger reads the ledgerstream ingestion
  cursor (upserted once per ledger), so it is the freshest tip
  signal available. /v1/ledger/stream is the SSE counterpart:
  it pushes a ledger_update event per new ledger (poll cadence
  ~2s) plus a keepalive refresh every ~10s so lag_seconds stays
  honest during an ingest stall — letting the status page render
  blocks arriving in real time instead of polling on a 30s timer.

Fixed

• Live-tail ingest lag sawtoothed 0→30s (#57). The SDK
  BufferedStorageBackend defaults RetryWait to 30s: once the
  indexer catches up to galexie's tip, the fetch worker requesting
  the next ledger object misses (galexie hasn't uploaded it yet)
  and sleeps a full 30s before re-checking — even though galexie
  uploads each LCM within ~5s of ledger close. The result was the
  indexer cursor advancing in bursts of ~5 ledgers every ~30s
  rather than tracking the tip continuously, so end-to-end ingest
  lag oscillated between ~2s and ~30s. Added
  ledgerstream.Config.LiveRetryWait — an unbounded-stream-only
  override (a missing object on a bounded range is still a hard
  error) — and set it to 3s for the galexie-live bucket in
  pipeline.LedgerstreamConfig. A caught-up worker now re-checks
  every 3s, collapsing the lag floor to ~3–8s.