Force reconnect on broken TCP connections

[mirek]

Issue we're experiencing with ioredis is described here.

Ie. in commonly used setup (ie. client connecting from container/common linux distros) kernel tcp defaults mean that abnormally terminated connections will be seen as ~15 minute failure with no activity on ioredis client (ie. TCP is connected but there is no activity from PoV of the client).

This situation can be controlled by setting net.ipv4.tcp_retries2 = 5 kernel param to lower downtime to roughly 1 minute – but I'd guess many teams are living with this issue possibly doing some hacks like recreating client if their custom heartbeat protocol fails etc.

Should ioredis be providing heartbeat/ping based auto reconnect, ie. something similar to what StackExchange.Redis did to address this issue?