A view that races a restart no longer fails twice

When a page was rendering at the very moment the part of the server serving it was being restarted — a rolling deploy, or a node's worker being recycled — the permission check behind the page kept asking the departing worker for its services, hit the closed door, and reported that as a rendering error. It then tried to show you an error message through the same closed door and failed a second time. You got neither the page nor the message, and two errors were filed for what was a routine restart.

The permission check now takes everything it needs from the worker once, when it starts, so a restart in the middle of a page can no longer trip it. And if a render does run into a worker that has already gone, that is treated as the restart it is: nothing is reported as an error, no message is attempted through the departed worker, and your page reconnects to the fresh worker by itself — the same way it already does when a worker announces its own restart.

Reconnecting…
The connection to the server was interrupted. Trying to restore it…
Trying again…
The connection could not be restored. Reloading the page…
The server was updated. Reloading the page to pick up the latest version.