feed went quiet for hours, socket still showed connected. staleness checks in prod?
running copytrade + automation across a few venues. one night a position sits wrong for hours and the bot does nothing about it.
blamed strategy logic first, then the signal path, both clean. turned out the market data websocket on one venue had died hours earlier. socket showed no error and still read as connected, messages just stopped.
venue numbers its book updates and i assumed that covered me. it only catches drops and dropped message leaves a gap in the sequence. dead feed sends nothing at all, so there's no gap and the check stays green while the book sits frozen. the only thing that flags it is time since the last message. copytrade made it worse, the mirror sizes off the master and once that state went stale it kept working from hours old numbers. no exceptions raised, i noticed when the accounts stopped matching.
what do you actually run in prod, app level pings with pong deadlines, comparing mids across venues?
mostly interested in one leg going dark while the rest look fine.