- sysctl: tcp_keepalive=60/10/3, conntrack_established=600s (was 5 days)
- nfqws/nfqws2: --ctrack-timeouts=30,30,600,30,30,30,30,30 in default strategy
- watchdog: probe_nfqueue checks nfqws/nfqws2 process health every 300s
- watchdog: heal_nfqueue_stopped restarts tachyon if DPI bypass processes die
This eliminates the need for manual DNS flush + device restart when DPI
starts blocking connections. Dead connections are now detected in ~90s
instead of hours, and stale conntrack entries expire in 10 min instead
of 5 days.
Dushnilin
404db8ae05