quazardous/qdadm @quazardous/qdcore@1.2.0

on GitHub

Minor Changes

• 05a334c: sse.idleTimeout — because a stream can die without anyone being told

  Measured against a proxy that cut the socket: EventSource went on reporting
  readyState OPEN with no error event, forever, while nothing arrived. A
  fetch reader on the same URL never had read() resolve either. Both APIs
  simply waited, and SSEBridge — which only listens for onerror — waited
  with them. Not the reported symptom of a stream reconnecting too often: a
  stream that stops for good and says nothing.

  Time since the last sign of life is the one signal that did not lie:

  sse: { url: '/events', idleTimeout: 45000 }

  Past the budget the bridge treats the stream as dead — sse:error,
  sse:disconnected, drop, and reconnect on its usual timer with a fresh ticket
  if you issue one.

  Off by default, because the right budget depends on how often your server
  speaks. Too short kills a quiet but healthy stream on a schedule.

  What counts as a sign of life differs by transport, and the difference is the
  browser's, not ours: an SSE heartbeat is a comment line, which EventSource
  consumes without surfacing anything. On the default transport the watchdog
  sees data frames only — budget for your data, not for your pings.

  Also new, and not wired to anything by default: SSEBridge now takes a
  transport. EventSourceTransport is the default and behaves exactly as
  before; FetchTransport reads the wire format itself, which lets it send
  headers — no credential in a query string, and none in an access log — and see
  a clean end-of-stream. It does not replace the watchdog: an abrupt socket cut
  is invisible to both.