HydraIssues

Head / Body

Known issues at event start

None.

Agent instructions

When hospitality flags a problem with this iPad:

1. Check GET /api/v1/heads/node-769ccedd and GET /api/v1/nodes/node-b961f1c8.
2. If stream_status=streaming with no active head: body self-heals (grace ~1m, GPU-mismatch ~3m). Force-stop only if stuck beyond 5 min.
3. If iPad stuck on pairing: check WireGuard on iPad (Settings → VPN) and ping body wireguard_ip from node-b961f1c8 exec.
4. Append findings under a new ### HH:MM — [summary] heading.
5. At every full hour and end of day: append an hourly/EOD summary of status, any incidents, and resolution.

Event log

10:25 — Event start

All systems nominal. ipad-head-3 streaming mercator-talks on chunky-turnip-23.

~14:30 — Body stuck in streaming state; iPad showing no body available

ipad-head-3 entered self-service mode (no active stream). chunky-turnip-23 was stuck in stream_status: streaming with no connected head. Only one body in district bxl1-test-3 — iPad could not discover any available body.

Body self-healed (stream_status returned to idle) within the grace window. iPad reconnected automatically.

~17:18 — Mercator fails to load after reconnect; Windows desktop visible

After reconnect, stream established successfully but Mercator Talks did not launch — Windows desktop visible on iPad.

Root cause: hydrabody ForceStop call to Sunshine API is missing Content-Type: application/json header. Sunshine rejects with HTTP 400. Hydrabody falls back to taskkill, but GPU stays at 98% briefly. New stream started during that window before Mercator could terminate cleanly.

Stream recovered on subsequent reconnect. Issue filed as #367 (hydrabody bug, one-line fix).

Secondary finding: hydraguard-air role assigned to chunky-turnip-23 but binary missing (windows-amd64 404 on releases). Non-critical, filed as #368.