Moving the heartbeat to after the initial load was my initial thought too but I don’t know the subtitles around the length of time the API loading takes on production servers vs the timeout of the node registration.
Regarding mutex while I concur that it does block across threads; however there is a time during the http connection where the mutex is unlocked, this could lead to concurrency issues where two threads are issuing a http request at exactly the same moment - they’ll be race condition and whichever completes first will succeed (as the nonce will/should only be valid once).
Agreed that there may be something dashboard side although no code for me to look through on that. I’ll try running dashboard in debug mode (if that mode exists?) and see if I can see anything.
Richard