tomspark wrote:
You've basically diagnosed it yourself, and it's not something you're doing wrong. The funnel setting isn't surviving SparkBox's nightly refresh — when the containers get recreated for updates/backups, the funnel falls back to tailnet-only and Seerr's forward drops with it. That lines up exactly with what you're seeing: works for a day, then needs a manual restart. Making that funnel stick across the nightly refresh is on us to fix, not you — I'm flagging it as a real gap to get it sorted. Until then, the reliable route is reaching Jellyfin and Seerr over your tailnet (through the Tailscale app) rather than the public funnel, since that side doesn't drop. I'll follow up here once there's a proper fix.
Chris wrote:
Coming back to this 51 days later than you deserved — but with a real fix rather than another workaround. The mechanism has been there a while: entries you declare in a serve.user.json file get merged into SparkBox's generated config on every container start, so the nightly recreate stops undoing them. What was still broken until today is that a core upgrade quietly deleted that file. Tom closed that hole in v1.6.484. sudo sparkbox upgrade Then declare your entries in /opt/sparkbox/state/tailscale/serve.user.json. One catch worth knowing: the dashboard owns tailnet 443, so give Jellyfin its own port there. — Chris