GraceDB uploads should be an async process
finalsink Should run synchronously (if not we have bigger problems), but right after clustering it can attempt a gracedb upload. That can sometimes take a second or two, and finalsink will block until its done.
So, if we see 2 triggers on the same compute node, and the 2nd passes trigger_control, we'd be adding a few seconds of latency.
The same goes for outputting coincs, launching snapshots, etc., though each of those Should be fast. (snapshots isn't always, but I'll make a separate issue for that).
Even if there isn't a followup gracedb upload, best not to hold up the pipeline by seconds. It takes time to work through the backlog.