On pod start the nzbget container raced gluetun: /etc/resolv.conf points at
10.128.0.1 (reachable only via the WireGuard tunnel), so for the ~20s gluetun
needs to establish the tunnel every DNS lookup from nzbget returned EAI_AGAIN.
Any in-queue download that had articles fetched during that window dropped
below the HealthCheck threshold (~97.9%) and was auto-cancelled — even items
that would otherwise complete (saw 97.6-97.8% health = "very nearly fine").
Override the nzbget container's entrypoint to poll DNS resolution and only
exec /init once it succeeds. That's the direct test of "tunnel is up + DNS
works", which is what nzbget needs.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
gilgamezh
2f24f75752