Architecture failsafe question

[satellite-no]

I had an event where my main sink went down and I lost logging to it and the buffer started writting to disk as expected. The unexpected thing however, was I lost my 'out-of-band' Prometheous Exporter data as well that had all of the Vector Internal node metrics. Why would I lose all of my internal metrics that go out a different sink?

Another qustion is we had one bad table definition that took down all logging for everything, is there a way to architect around this so we dont lose all logging? This seems very dangoues that 1 bad sinkcould take down TB of logging?