DanivosYoun 855b2b272f fix(websocket): detect dead/half-open control connection (read deadline + protocol ping) ...

The control websocket had no read deadline and only sent application-level
"newt/ping" JSON (never protocol PING frames), so a half-open connection — for
example a cloud load balancer keeping the socket open after the backend API
server restarted — was never detected. ReadMessage blocked forever while the
buffered ping writes kept succeeding, so newt stayed "connected" to a dead peer
and only recovered after a manual restart.

- Arm a read deadline (pongWait = 2x pingInterval, min 20s) and refresh it on
  every pong and every inbound message.
- Send a protocol-level PING alongside the existing app ping, so a compliant
  server replies with a PONG that refreshes the read deadline.
- On ping-write failure, close the connection and let the read pump perform a
  single reconnect (previously sendPing and the read-pump defer could both
  trigger a reconnect, racing two connections).
- Give each connection a lifecycle channel so its ping monitor stops when the
  connection ends (previously one ping-monitor goroutine leaked per reconnect).

2026-06-22 11:51:03 +09:00

..

client.go

fix(websocket): detect dead/half-open control connection (read deadline + protocol ping)

2026-06-22 11:51:03 +09:00

config_test.go

…

config.go

Use logger not printf

2026-06-04 16:57:51 -07:00

types.go

Add sighup and block flag to block connections

2026-05-28 15:38:09 -07:00