cache-mesh@0.2.1

Patch Changes

• 77eb949 Thanks @ridakaddir! - Ship the code review fixes landed in #2.

  Correctness & availability:

  • StoreWrapper.get() / has() are now O(1) when no tombstone has expired; previously every read scanned the full meta Map.
  • /sync/snapshot honours HTTP backpressure, avoiding buffer blow-up on large caches with slow peers.
  • Unhandled emit('error') on the Coordinator no longer crashes the host process.
  • NDJSON parser rejects lines above 4 MiB and surfaces malformed JSON via a typed NdjsonParseError.
  • bootstrap:finished now carries outcome: 'ok' | 'empty-cluster' | 'all-failed' so callers can tell success from silent failure.

  Robustness:

  • Per-peer outbox rewritten as a circular buffer with a droppedCount() signal; SyncClient logs drops.
  • coordinator.start() rolls back partially-started resources on failure.
  • Discovery listeners are detached on stop() to prevent leaks across restarts.
  • Remote ops whose HLC physical time is > 5 minutes ahead of local wall clock are rejected, preventing clock-poisoning.
  • tombstoneTtlMs JSDoc documents the partition-duration constraint.

  Polish:

  • Environment variable convention unified to CACHE_MESH_* across README and examples (CACHE_SYNC_* removed). Wire headers (x-cachesync-*) unchanged for on-wire compatibility.
  • outboxCapacity validated as a positive integer.
  • onOp failures return 500 so peers requeue instead of treating a broken apply as delivered.