yggdrasil-jumper: exits ~10 s after start with yggdrasil 0.5.15 (admin socket timeout vs 10 s query delay) #31258

Package Name

yggdrasil-jumper

Maintainer

@one-d-wide

OpenWrt Version

master (yggdrasil 0.5.15-r1, yggdrasil-jumper 0.3.1-r1). Reproduced on 25.12.6 with yggdrasil 0.5.15 built from the master recipe in the 25.12.5 SDK.

OpenWrt Target/Subtarget

mediatek/filogic

Steps to Reproduce

1. Install yggdrasil 0.5.15 and yggdrasil-jumper 0.3.1, and configure an interface (here ygg0) with proto 'yggdrasil' and jumper_enable '1'.
2. Bring the interface up: ubus call network.interface.ygg0 up
3. Wait 15 s, then run pidof yggdrasil-jumper and logread -e ygg0-jumper.

Actual Behaviour

The jumper connects to the admin socket and exits about 10 s later:

01:39:55 ygg0-jumper:  INFO Admin API: Connected to unix:///tmp/yggdrasil/ygg0.sock
01:40:05 ygg0-jumper: ERROR Admin API watcher: Failed to query admin api: Broken pipe (os error 32)

After that pidof yggdrasil-jumper prints nothing. The interface and its peers keep working, but there is no jumper until the next setup of the interface.

Cause: Yggdrasil 0.5.15 added "Admin socket reads now have a 1MB limit and a 10 second timeout" to its changelog, so the daemon closes an admin connection that sends no request for 10 s. Jumper 0.3.1 keeps one connection open and sends a request every yggdrasilctl_query_delay, which defaults to 10 s. The daemon closes the connection just before the next request, and 0.3.1 has no reconnect: yggdrasil_admin_reconnect / --reconnect only arrived in jumper 0.4.0, and is off by default there too.

Workaround: shorten the query delay through the package's existing jumper_config option, then set the interface up again. The jumper has then stayed up on two routers.

uci set network.ygg0.jumper_config='yggdrasilctl_query_delay = 5.0'
uci commit network
ubus call network.interface.ygg0 down; ubus call network.interface.ygg0 up

Expected Behaviour

The jumper keeps running with the yggdrasil version shipped in the same feed. The package could do this in several ways:

• update to jumper 0.4.x and enable yggdrasil_admin_reconnect;
• write a query delay below 10 s into the generated config (proto_yggdrasil_generate_jumper_config in net/yggdrasil/files/yggdrasil.sh);
• have netifd restart the jumper.

Additional Information

Related: #31247 covers the jumper start-up in the same proto_yggdrasil_setup(), where a duplicate jumper appears after a quick re-setup and 2&>1 sends stderr to a file.

Confirmation Checklist

• The package is maintained in this repository.
• I understand that issues related to the base OpenWrt repository or LuCI repository will be closed.
• I am reporting an issue for OpenWrt, not an unsupported fork.