We're experiencing this exact same issue, and it's very negatively affecting our API service. I thought Apple would fix it shortly after shutting down DarkSky, but it's still as bad of a problem, if not worse. I'm currently doing workarounds such as increasing the number of retries, and allowing manual retries as well, but this really needs to be fixed by Apple.
How can we get Apple to prioritize this more highly? It seems like there's been no action taken yet.