A data source that fails to load no longer freezes its node

When a node's data failed to load, it was supposed to say so immediately: every request got a clear "initialization failed — here is why" answer within milliseconds, and the page showed the real reason.

Occasionally it did the opposite. The node recorded the failure internally and then went silent: requests were held back waiting for a load that had already given up, and callers waited out their whole budget — up to 30 seconds — before getting a generic timeout with no explanation. Whether it happened came down to which of two things finished first on a busy machine, so the same node could behave correctly one moment and freeze the next.

The two are no longer in a race. A data load that fails now always reaches the code that reports it, so the node reports the error immediately every time, and the message you see names the actual problem instead of a timeout.