A node restarting no longer strands the page that just opened it

Nodes restart routinely — after a republish, an idle deactivation, or a self-heal — and that normally passes unnoticed. But if you happened to open or save something in the split second a node was restarting, and nothing on your screen had loaded from it yet, the connection was quietly abandoned: the view sat empty, or the save spun for thirty seconds and then reported a timeout. Trying again worked, which is why this mostly looked like a random hiccup.

The connection now waits for the restart to finish before asking again, so the work simply lands a moment later instead of failing. Installs are the most visible beneficiary — one could previously report zero items written when a node it was writing to restarted mid-run.