Is there any tools that does log/trace capture on error conditions? I.e. we capture all local events, but only upload them when something meaningful happens, like the server crashed / requests are returning 5xx.

I love this idea in principle, but in practice I would guess it means one of two sub-optimal things: either the node caches them for a window of time, in order to know whether to really transmit them, or the logs are mutated post-delivery as kind of a "tiny expiry"

Everything else I could write is just turning various trade-off knobs, which is why I'd guess you haven't seen an out-of-the-box offering that does what you're describing. There's not just one solution to it that would be reasonable for all audiences